MNRAS 000, 1–18 (2024) Preprint 19 March 2024 Compiled using MNRAS LATEX style file v3.

EDGE – Dark matter or astrophysics? Breaking dark matter heating

degeneracies with H i rotation in faint dwarf galaxies
Martin P. Rey,1★ Matthew D. A. Orkney,2 Justin I. Read,3 Payel Das,3 Oscar Agertz4 , Andrew Pontzen5 ,
Anastasia A. Ponomareva1 , Stacy Y. Kim3 and William McClymont 3,6,7
1 Sub-department of Astrophysics, University of Oxford, DWB, Keble Road, Oxford OX1 3RH, UK
2 Institut de Ciencies del Cosmos (ICCUB), Universitat de Barcelona (IEEC-UB), Martí i Franquès 1, E08028 Barcelona, Spain
3
arXiv:2309.00041v2 [astro-ph.GA] 16 Mar 2024

Department of Physics, University of Surrey, Guildford GU2 7XH, UK

4 Lund Observatory, Division of Astrophysics, Department of Physics, Lund University, Box 43, SE-221 00 Lund, Sweden
5 Department of Physics and Astronomy, University College London, London WC1E 6BT, UK
6 Kavli Institute for Cosmology, University of Cambridge, Madingley Road, Cambridge, CB3 0HA, UK
7 Cavendish Laboratory, University of Cambridge, Madingley Road, Cambridge, CB3 9BB, UK

Submitted to MNRAS

ABSTRACT
Low-mass dwarf galaxies are expected to reside within dark matter haloes that have a pristine, ‘cuspy’ density profile within their
stellar half-light radii. This is because they form too few stars to significantly drive dark matter heating through supernova-driven
outflows. Here, we study such simulated faint systems (104 ≤ 𝑀★ ≤ 2×106 M⊙ ) drawn from high-resolution (3 pc) cosmological
simulations from the ‘Engineering Dwarf Galaxies at the Edge of galaxy formation’ (EDGE) project. We confirm that these
objects have steep and rising inner dark matter density profiles at 𝑧 = 0, little affected by galaxy formation effects. But five dwarf
galaxies from the suite also showcase a detectable H i reservoir (𝑀H i ≈ 105 − 106 M⊙ ), analogous to the observed population
of faint, H i-bearing dwarf galaxies. These reservoirs exhibit episodes of ordered rotation, opening windows for rotation curve
analysis. Within actively star-forming dwarfs, stellar feedback easily disrupts the tenuous H i discs (𝑣 𝜙,𝑔 ≈ 10 km s−1 ), making
rotation short-lived (≪ 150 Myr) and more challenging to interpret for dark matter inferences. In contrast, we highlight a
long-lived (≥ 500 Myr) and easy-to-interpret H i rotation curve extending to ≈ 2 𝑟 1/2,3D in a quiescent dwarf, that has not formed
new stars since 𝑧 = 4. This stable gas disc is supported by an oblate dark matter halo shape that drives high-angular momentum
gas flows. Our results strongly motivate further searches for H i in rotation curves in the observed population of H i-bearing
low-mass dwarfs, that provide a key regime to disentangle the respective roles of dark matter microphysics and galaxy formation
effects in driving dark matter heating.
Key words: methods: numerical – galaxies: structure – galaxies: evolution – dark matter

1 INTRODUCTION Armengaud et al. 2017; Rogers & Peiris 2021) or their combina-
tion (e.g. Nadler et al. 2021b; Enzi et al. 2021) all point to dark
The existence of a significant amount of dark matter in our Universe
matter being a cold (i.e. non-relativistic at the time of decoupling)
is firmly established, with its gravitational influence leaving distinct
collisionless, particle. However, this stills leaves plenty of possible
signatures on the cosmic microwave background (e.g. Planck Col-
options for the physical nature of the constituent sourcing the dark
laboration et al. 2020), the large-scale distribution of galaxies (e.g.
matter gravitational field (e.g. supersymmetric weakly interacting
Alam et al. 2021), and the dynamics of baryonic tracers in galaxies
massive particles, sterile neutrinos, axions, etc; see Bertone et al.
and galaxy clusters (e.g. Zwicky 1933; Rubin et al. 1980; Clowe
2005; Bertone & Tait 2018 for reviews).
et al. 2006). But the microphysical nature of dark matter and its di-
rect detection remains elusive, despite extensive efforts in the last Galactic rotation curves are one of the first historical probes of
decade (see e.g. Schumann 2019 for a review). This calls for a wide dark matter and continue to play a key role in the effort to narrow
and thorough scan of parameter space to robustly remove alterna- down the available parameter space of models (e.g. Rubin & Ford
tives, motivating complementary efforts across disciplines (Bertone 1970; Rubin et al. 1980; van Albada et al. 1985; de Blok & Bosma
& Tait 2018). 2002; Oh et al. 2011, 2015; Lelli et al. 2016; Posti et al. 2019;
The latest data from galaxy counts (e.g. Nadler et al. 2021a), stel- Mancera Piña et al. 2020). In particular, rotation curves and H i
lar stream gaps (e.g. Banik et al. 2021), strong lensing (e.g. Gilman kinematics of small dwarf galaxies are particularly powerful. They
et al. 2020; Hsueh et al. 2020), the Ly 𝛼 forest (e.g. Iršič et al. 2017; can be used to determine the dark matter halo masses hosting small
dwarf galaxies, directly constraining the low-mass end of the galaxy-
halo connection and dark matter models that suppress small-scale
★ E-mail: [email protected] power in the cosmological power spectrum (e.g. a warm or wave dark

© 2024 The Authors

2 M. P. Rey et al.
matter; Polisensky & Ricotti 2011; Anderhalden et al. 2013; Kennedy Excitingly, a growing number of isolated, gas-rich faint dwarfs
et al. 2014; Read et al. 2017; Nadler et al. 2021a; Yasin et al. 2023; have recently been reported, showcasing small but detectable H i
Sardone et al. 2023). Rotation curves are also sensitive to the structure reservoirs (𝑀H i ≈ 105 − 106 M ⊙ ) that matches a faint stellar com-
of the inner gravitational potential and can be used to infer the dark ponent (104 ≤ 𝑀★/M ⊙ ≤ 106 ; Irwin et al. 2007; Cole et al. 2014;
matter distributions and density profiles in dwarf galaxies (e.g. Flores McQuinn et al. 2015, 2020, 2021; Sand et al. 2015; Adams & Oost-
& Primack 1994; Moore 1994; de Blok & Bosma 2002; Oh et al. erloo 2018; Brunker et al. 2019; Janesh et al. 2019; Hargis et al.
2011, 2015; Ott et al. 2012; Iorio et al. 2017). This in turn provides 2020; Bennet et al. 2022; Rhode et al. 2023). Such low stellar masses
constraints on mechanisms heating dark matter, either dynamically (and thus low galaxy formation effects) and the presence of H i (and
or through microphysical particle interactions (e.g. annihilation or thus of a dynamical tracer) could provide precisely the rotation curves
self-interactions). needed to cleanly separate dark matter models from galaxy formation
Both of these features have historically garnered significant inter- effects. Further, these objects are typically isolated ‘field’ dwarfs, re-
est from the community because, at face value, they are discrepant moving the need to model environmental effects from more massive
with predictions assuming pure cold dark matter (CDM; see Pontzen hosts.
& Governato 2014; Bullock & Boylan-Kolchin 2017; Sales et al. However, a key puzzle remains before we can leverage this popu-
2022 for reviews). CDM-only structure formation predicts many lation of dwarf galaxies as a dark matter probe: none of them so far
more bound dark matter subhaloes than observed satellite galaxies shows evidence for clear, ordered rotation in their H i gas that can
around the Milky Way and other nearby spiral galaxies – the ‘miss- be easily exploited for dynamical modelling (Bernstein-Cooper et al.
ing satellite problem’ (e.g. Moore et al. 1999; Klypin et al. 1999). 2014; Adams & Oosterloo 2018; McQuinn et al. 2021). It remains
These same simulations predict centrally divergent dark matter den- unknown whether this is due to unfortunate inclination in the few ex-
sity profiles inside dwarf galaxies, ‘cusps’, whereas observations amples observed (i.e. near face-on orientations), to the observational
favour lower density ‘cores’ – the ‘cusp-core problem’ (e.g. Flores & challenges associated with working with such small galaxies (e.g.
Primack 1994; Moore 1994; de Blok & Bosma 2002; Oh et al. 2011, Read et al. 2016b; Verbeke et al. 2017; Oman et al. 2019; Downing
2015; Iorio et al. 2017). Both problems can be solved by moving & Oman 2023 for discussions), or to an intrinsic lack of ordered
beyond the CDM assumption. For example, the cusp-core problem rotation in the H i gas altogether at this mass scale.
can be mitigated by making dark matter self-interacting (e.g. Burkert In this paper, we address this puzzle using a suite of high-resolution
2000; Spergel & Steinhardt 2000) or fuzzy (e.g. Schive et al. 2014; cosmological ‘zoomed’ simulations of faint dwarf galaxies from the
Veltmaat et al. 2018; Nori & Baldi 2021), while satellite numbers EDGE project (introduced in Agertz et al. 2020). Our sample of
can be reduced by suppressing small-scale cosmological power (e.g. simulated galaxies matches the observed population of isolated faint
Boehm et al. 2014; Vogelsberger et al. 2019). H i-rich dwarfs, in both stellar masses, H i masses and star formation
However, these discrepancies can also be explained by a careful activity (or lack thereof; Rey et al. 2019, 2020, 2022; Section 2).
modelling of the physics of galaxy formation. In the case of the In Section 3, we extract their gas and H i kinematics, highlighting
missing satellite problem, the solution involves a mix of accounting multiple examples of rotationally supported gas kinematics and H i
for observational completeness (e.g. Kim et al. 2018), star forma- rotation curves. This includes short-lived H i discs rapidly dispersed
tion becoming inefficient in low mass haloes (e.g. Efstathiou 1992; by the energy input from massive stars, but also a long-lived example
Somerville 2002; Sawala et al. 2016; Read & Erkal 2019) and the with near-circular rotation that could easily be modelled by standard
tidal destruction of satellites on plunging orbits (e.g. Read et al. mass-modelling tools (Section 4). We discuss the physical drivers of
2006; Garrison-Kimmel et al. 2017). In the case of the cusp-core our results and their significance for future observational campaigns
problem, it can be solved by dark matter being dynamically heated in Section 5.
during galaxy formation via repeated gas inflows and outflows (e.g.
Navarro et al. 1996; Read & Gilmore 2005; Pontzen & Governato
2012), or via dynamical perturbations induced by massive clumps
2 THE KEY REGIME OF FAINT AND H i -RICH DWARFS
or companions (e.g. El-Zant et al. 2001; Romano-Díaz et al. 2009;
Goerdt et al. 2010; Nipoti & Binney 2015; Orkney et al. 2021). There We use the suite of faint (104 ≤ 𝑀★ ≤ 2 × 106 M⊙ ) simulated
is now compelling observational evidence that ‘dark matter heating’ dwarf galaxies presented in Rey et al. (2022), specifically focusing
occurred in nearby dwarfs (e.g. Read et al. 2019; Bouché et al. 2022; on the subset of five H i-bearing dwarfs (105 ≤ 𝑀H i ≤ 106 M⊙ ).
De Leo et al. 2023). Next, we briefly summarize how each simulated galaxy is evolved to
This makes testing dark matter models with rotation curves more 𝑧 = 0 using cosmological, zoomed hydrodynamical simulations (see
ambiguous, as the effects of alternative dark matter models become Agertz et al. 2020; Rey et al. 2020 for more in-depth descriptions)
degenerate with the physics of galaxy formation which remains chal- and the characteristics of the simulated suite (see also Rey et al. 2019,
lenging to model from first principles (see Somerville & Davé 2015; 2020, 2022; Orkney et al. 2021, 2023).
Naab & Ostriker 2017 for reviews). This motivates us to find ‘clean’ All galaxies are evolved to 𝑧 = 0 using cosmological zoomed, hy-
regimes to test models – galaxies in which the rotation curve data are drodynamical simulations with the adaptive mesh refinement ram-
straightforward to interpret, and where dark matter and astrophys- ses code (Teyssier 2002). The mass resolution inside the galaxy’s
ical models make testable predictions with minimal overlap. The Lagrangian region is enhanced using the genetIC software (Stopyra
best candidates for this are the smallest dwarf galaxies, where low et al. 2021) to reach 𝑚 DM = 960 M⊙ , while the hydrodynamical
stellar masses leave little opportunity for star formation and galaxy refinement strategy ensures a spatial resolution of 3 pc across the
formation effects to impact the inner dark matter density profile (e.g. galaxy’s interstellar medium (ISM; Agertz et al. 2020). The cosmo-
Teyssier et al. 2013; Di Cintio et al. 2014; Chan et al. 2015; Tollet logical streaming of the Lagrangian patch of each galaxy is zeroed to
et al. 2016; Read et al. 2019; Lazar et al. 2020; Orkney et al. 2021). reduce advection errors (Pontzen et al. 2021). We follow the forma-
But these ‘ultra-faint’ dwarfs are typically devoid of gas (Geha et al. tion of stars and the injection of energy, momentum, mass, and metals
2012; Putman et al. 2021) and thus unsuitable for rotation curve from asymptotic giant branch (AGB) stars, Type-II and Type-Ia su-
analysis. pernovae (SNII, SNIa) according to Agertz et al. (2020). We track

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 3
the cooling of primordial and metal-enriched gas using equilibrium the physical mechanisms affecting the timing of star formation reig-
thermochemistry (Courty & Alimi 2004), accounting for on-the-fly nition).
self-shielding (Aubert & Teyssier 2010; Rosdahl & Blaizot 2012) These differences in star-formation activity are directly reflected
and heating from a spatially uniform, time-dependent UVB (updated in our simulated dwarfs’ H i properties. Star-forming dwarfs show
from Haardt & Madau 1996; see Rey et al. 2020 for further details). strongly time varying, asymmetric H i reservoirs that are often spa-
To derive H i distributions, we evaluate the code’s internal cooling tially offset from their stellar body (Rey et al. 2022). Quiescent
function at every spatial position of the simulation and compute the dwarfs, by contrast, show more stable, more spherical and more
neutral hydrogen fraction (Rey et al. 2022). We track dark matter aligned H i contents over time (Rey et al. 2022). As we will see next,
haloes over time using the hop halo finder (Eisenstein & Hut 1998) these distinctions are also reflected in the stability and structure of
and construct merger trees using the pynbody and tangos libraries their gas and H i kinematics.
(Pontzen et al. 2013; Pontzen & Tremmel 2018). We centre on our Furthermore, all of our simulated dwarf galaxies exhibit steep
galaxies using the shrinking sphere algorithm (Power et al. 2003) on and rising dark matter density profiles around 𝑟 1/2,3D . This is first
the dark matter and shift the velocity frame to put the central 1 kpc at shown in Orkney et al. (2021) using higher-resolution re-simulations
rest. We interpolate a single stellar population model (Girardi et al. (𝑚 DM = 120 M⊙ ) of a subset of the galaxies studied here. This con-
2010) over a grid of ages and metallicities to obtain the luminosities clusion also holds at the resolution of this work (𝑚 DM = 960 M⊙ ),
of individual stellar particles and compute the 3D stellar half-light with Figure 1 showing the spherically averaged dark matter density
radius, 𝑟 1/2,3D . profiles at 𝑧 = 0 using 100 log-spaced bins. All profiles are con-
The EDGE simulated suite consists of ten low-mass dwarf galaxies sistent with an increasing dark matter density towards the centre,
(𝑀★ ≤ 2×106 M⊙ ) hosted in dark matter haloes with 109 ≤ 𝑀200 ≤ with a ‘cuspy’ logarithmic slope (≈ −1) until the limited resolution
3 × 109 M⊙ at 𝑧 = 0 (Rey et al. 2022). At this mass-scale, all of the simulations starts affecting the dynamics of dark matter par-
of our galaxies see their star formation truncated at high redshift ticles (grey band; see Orkney et al. 2021, appendix B for further
(𝑧 ≥ 4) following cosmic reionization, as their potential wells are discussion).
then too shallow to accrete gas from the intergalactic medium (see Figure 1 emphasizes that we have reached a critical regime. At
e.g. Efstathiou 1992; Gnedin 2000; Hoeft et al. 2006; Okamoto et al. this galactic mass scale, dynamical effects and star formation-driven
2008; Noh & McQuinn 2014 for further discussions). outflows can lower central dark matter densities in the very centre
Five of our dwarf galaxies assemble little dynamical mass at late of dwarfs, but are inefficient at forming large (i.e. ≈ 𝑟 1/2,3D ) and
times (i.e. after reionization) and have vanishing gas and H i contents flat (i.e. constant-density) dark matter cores (see Orkney et al. 2021
at 𝑧 = 0. Conversely, five others grow enough at late times to start for further discussion). Any observational evidence for 𝑟 1/2,3D -sized
re-accreting gas from the hot intergalactic medium and eventually dark matter cores in such faint dwarfs (e.g. Amorisco 2017; Sanders
host a detectable H i reservoir at 𝑧 = 0 (Rey et al. 2020, 2022; see et al. 2018; Malhan et al. 2022) thus becomes increasingly difficult to
also Ricotti 2009; Benítez-Llambay et al. 2015; Fitts et al. 2017; explain through purely astrophysical effects and could rather point to
Jeon et al. 2017; Ledinauskas & Zubovas 2018; Benitez-Llambay new dark matter physics (see e.g. Orkney et al. 2022 for further dis-
& Frenk 2020; Benitez-Llambay & Fumagalli 2021; Pereira Wilson cussion). The gas reservoirs of faint and H i-bearing dwarfs provide
et al. 2023 for further discussion of this re-accretion mechanism). a unique opportunity to obtain such observational insights provided
The five H i-bearing objects are the focus of this study. With 104 ≤ that H i kinematics can be harnessed to infer the structure of host
𝑀★ ≤ 2×106 M⊙ and 105 ≤ 𝑀H i ≤ 106 M⊙ , they provide excellent dark matter haloes which we now quantify.
simulated analogues to the observed population of low-mass, H i-
bearing dwarfs in the 𝑀★ − 𝑀H i plane (Rey et al. 2022, fig. 2).
Furthermore, three out of five galaxies re-accreted their gas reser-
3 DIVERSE AND VARIABLE H i KINEMATICS IN FAINT
voirs early enough to re-ignite star-formation several billion years
DWARFS
ago. Since their re-ignition, these galaxies have been forming stars
with star formation rates (SFRs) ≈ 10 −5 M⊙ yr −1 averaged over sev- Figure 2 illustrates the diversity of gas and H i kinematics found
eral billion years (Rey et al. 2020, fig.1), with instantaneous peaks up across our sample of simulated gas-rich faint dwarfs. The top panels
to SFR ≈ 10 −4 M⊙ yr −1 (Rey et al. 2020, fig.7). These SFRs are sim- show an actively star-forming dwarf; the bottom a quiescent dwarf;
ilar to those measured in observed star-forming, low-mass dwarfs. For each at three output times spanning the last billion years of their
example, resolved colour-magnitude diagrams infer SFRs of a few evolution (selected to be the same output times as in Rey et al. 2022,
10 −5 M⊙ yr −1 averaged over several billion years in Leo T (Clemen- fig. 1). The maps plot the gas velocity weighted by H i mass along
tini et al. 2012; Weisz et al. 2012), Leoncino (McQuinn et al. 2021), the same line of sight in all panels. Contours show constant 1018 ,
and Antlia B (Hargis et al. 2020). The H 𝛼 detection of the single 1019 , 1020 and 1021 cm −2 H i column densities (black) and 𝑟 1/2,3D
H ii region of Leo P similarly implies SFR≈ 4 × 10 −5 M ⊙ yr −1 (blue circles).
(Rhode et al. 2013; McQuinn et al. 2015), while the presence of blue Focusing first on the star-forming dwarf (top panels), we recover
helium-branch stars highlight recent, short peaks of ≈ 10 −4 M⊙ yr −1 H i distributions that are strongly and rapidly varying in time (see Rey
in Coma P (Brunker et al. 2019). Overall, our simulated galaxies are et al. 2022 for an in-depth quantification). In these low-mass systems,
an excellent match to observed SFRs, H i contents, stellar masses and stellar feedback drives asymmetric and disturbed H i morphologies
stellar metallicities (Collins & Read 2022, fig.5) of faint dwarfs, mak- (top, middle), that are often offset from the galaxy’s stellar distribu-
ing them an ideal platform to predict H i kinematics in this regime. tion (e.g. top, left), and that can become temporarily unobservable
In addition to the three regularly star-forming dwarfs, another following powerful outflow and heating (top right; 𝑀H i marked on
galaxy of our sample re-ignited star formation just before 𝑧 = 0 each panel). This dynamic and time-varying behaviour is reflected in
(at 𝑧 = 0.03, 500 Myr ago), after several billion years of quiescent the gas kinematics. Over one billion years, gas travels in bulk flows
but gas-rich evolution. The last galaxy is yet to reignite star forma- of opposite directions within 𝑟 1/2,3D (top, left and right), but also
tion despite hosting a significant gas reservoir (see Rey et al. 2020; exhibits a snapshot of potential rotation with a disturbed but apparent
Benitez-Llambay & Fumagalli 2021; Pereira Wilson et al. 2023 for gradient in line-of-sight velocity around 𝑟 1/2,3D (top, centre).

MNRAS 000, 1–18 (2024)

4 M. P. Rey et al.
the lifetime of SNII (both ≈ 10 Myr), particularly for actively star-
9 cus forming objects. Unfortunately, this time-scale is much shorter than
10 p the cadence with which we save simulation outputs for each galaxy
(≈ 100 Myr). Combined with the Eulerian nature of our simulation
core code that limits the tracking of gas over time, this makes it challenging
to establish causal evolutionary trends from one snapshot to the next
(M kpc 3 )

r1/2, 3D (e.g. in Figure 2, the top panels are difficult to relate to one another as
multiple star formation bursts and H i re-accretion has occurred be-
108 tween them). Similarly to Rey et al. 2022, we thus adopt a statistical
approach, treating each snapshot as an independent realization of the
star forming cycle, flagging times of potential organized gas rotation
DM

across each dwarf’s history in subsection 4.1, and then study the H i
kinematics at those times in more details in subsections 4.2 and 4.3.
Halo 600 In future work, we will alleviate this issue leveraging algorithms that
7 Halo 600 GM: Delayed mergers
10 Halo 605
allow high-cadence tracking of gas dynamics (Cadiou et al. 2019) to
connect star formation activity and H i properties more causally (S.
Halo 624 Hutton et al. in preparation).
Halo 624 GM: Higher final mass
1
10 100
Radius (kpc) 4.1 Existence and prevalence of gas rotation
We start by computing, for each simulated snapshot, profiles of the
Figure 1. Dark matter density profiles across our suite of simulated, H i- tangential gas velocity 𝑣 𝜙,𝑔 , the circular velocity 𝑣 circ , the 3D isother-
bearing faint dwarf galaxies. At this galactic mass scale, dynamical effects and mal sound speed 𝑐 𝑠 and the 3D gas turbulent velocity 𝜎turb, g to
supernova-driven outflows naturally arising in ΛCDM cosmologies can re- quantify rotational, gravitational, thermal, and turbulent support re-
duce central dark matter densities but are inefficient at forming large (𝑟1/2,3D - spectively (see Appendix A for formal definitions). We then compute
sized; grey box) and flat dark matter cores (indicative profile slopes marked
the projected radial profiles of 𝑣 𝜙,𝑔 , 𝑐 𝑠 and 𝜎turb, g viewed face-on
in grey). Inferring the structures of dark matter haloes hosting these faint
dwarfs, for example through H i rotation (Figure 2), thus holds great promises
(i.e. in the plane of the disc) in 100 bins linearly spaced between
to distinguishing whether galaxy formation effects or new dark matter in- 0 and 2 kpc√︃and construct the effective velocity dispersion of the
teractions drive dark matter heating in dwarfs. Interpreting the flattening of gas 𝜎eff = 𝑐2𝑠 + 𝜎turb,2
g
. The 3D 𝑣 circ profile is derived from the
profiles at small radii (within marked grey box) is compromised by the limited full gravitational potential in the same radial range, sourced by the
resolution of the simulation.
combination of dark matter, gas and stars (but strongly dominated by
the dark matter at all radii for these faint objects).
In contrast, the quiescent system (bottom row) shows much more Figure 3 shows the evolution of 𝑣 𝜙,𝑔 /𝑣 circ and 𝑣 𝜙,𝑔 /𝜎eff evaluated
stable gas content, slowly accumulating H i gas over the last billion at 150 pc, where the highest column density H i is most often found
years (growing 𝑀H i at constant 𝑀★; see also Rey et al. 2020, 2022). (Rey et al. 2022). We only show the late-time evolution of these
Furthermore, this H i reservoir shows a distinctly flattened morphol- dwarfs (𝑧 ≤ 2), that is when they host detectable H i (see Rey et al.
ogy, with a positive-to-negative line-of-sight velocity gradient across 2022, fig. 1 for the time evolution of 𝑀H i over time of each of these
𝑟 1/2,3D at all time stamps. galaxies) and omit their earlier phase where saved simulation outputs
These two examples summarize well the more complete and quan- are sparser and miscentring due to mergers make 𝑣 𝜙,𝑔 and 𝑣 circ even
titative investigation presented in the next section. Star-forming low- noisier. Trends in Figure 3 are qualitatively unchanged if measuring
mass dwarfs host short-lived instances of H i rotation across their velocities at 100, 200 pc or at each galaxy’s 𝑟 1/2,3D instead.
evolution, but the small and tenuous discs are rapidly disrupted by Focusing first on 𝑣 𝜙,𝑔 /𝑣 circ (left-hand panels), we recover that gas
the energy input from newborn massive stars. Quiescent systems have kinematics are strongly variable in time, without clear evolutionary
more stable H i reservoirs and kinematics, increasing (but not guar- trends for star-forming low-mass dwarfs (blue) after the re-ignition
anteeing) their chances to host organized and long-lived H i rotation of their star formation (marked by stars in Figure 3). This is expected
ideal for inferences of the structure of their host dark matter halo. as stellar feedback efficiently disrupts the ISM in these shallow po-
tential wells (𝑣 circ ≈ 10 km s −1 at 𝑟 1/2,3D ). Nonetheless, some peaks
approach 𝑣 𝜙,𝑔 /𝑣 circ ≈ 1 (grey line in Figure 3) indicating potential
short-lived episodes where the rotational velocity 𝑣 𝜙,𝑔 is close to
4 ROTATING H i DISCS IN FAINT DWARFS AND THEIR
equilibrium with 𝑣 circ sourced by the underlying gravitational po-
PHYSICAL DRIVERS
tential. Furthermore, during these episodes, rotational motions can
We now aim to gain more quantitative insights into the gas rotational dominate over turbulent support, with 𝑣 𝜙,𝑔 /𝜎eff ≥ 1 (right-hand
support of our galaxies. In particular, we wish to (i) establish whether panels).
organized gas rotation can dominate thermal and turbulent motions, To quantify this further, we extract all time instances when 0.75 ≤
and thus be clearly identified observationally; (ii) test whether this 𝑣 𝜙,𝑔 /𝑣 circ ≤ 1.25 (i.e. loosely bracketing gas in circular rotation,
rotation is close to circular and in equilibrium, and thus easy to re- acknowledging that 𝑣 𝜙,𝑔 is not yet corrected for pressure support;
late to the host gravitational potential; and (iii) gain insights into the see Section 4.2 and Appendix A) and 𝑣 𝜙,𝑔 /𝜎eff ≥ 0.75 (i.e. rotation
prospects of characterizing such rotating gas with radio interferom- loosely dominating over thermal and kinetic turbulence). These cuts
eters. should not be interpreted quantitatively, but rather as helpful to flag
Gas contents and kinematics in our galaxies can be strongly vary- the likely presence of a galactic gas disc in the noisy kinematics of our
ing on time-scales comparable to the local dynamical times and to sensitive objects. We mark these times with diamonds in Figure 3,

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 5
t = 12.9 Gyr t = 13.2 Gyr t = 13.8 Gyr
20
M = 3.5 × 105 M M = 3.6 × 105 M M = 3.5 × 105 M
MHI = 4.8 × 105 M MHI = 1.2 × 106 M MHI = 4.9 × 103 M
Star-forming faint dwarf
Halo600

10
r1/2

vlos (km s 1 )
1 kpc 1 kpc 1 kpc
0
M = 5.6 × 105 M M = 5.6 × 105 M M = 5.6 × 105 M
MHI = 2.2 × 105 M MHI = 3.3 × 105 M MHI = 4.7 × 105 M
Gas-rich relic
Halo624

10

1 kpc 1 kpc 1 kpc

20
Figure 2. Maps of the gas velocity, H i-mass weighted along the same line of sight for a star-forming (top) and quiescent (bottom) simulated dwarf. From left
to right, we show the galaxies at three different times. Actively star-forming dwarfs (top) show irregular and time-varying H i distributions (the black contours
show 1018 , 1019 , 1020 and 1021 cm −2 column densities) following their cycle of gas accretion and stellar feedback (Rey et al. 2022). This dynamic behaviour is
reflected in their gas kinematics, which are often disturbed and show rotational support in short-lived episodes (e.g. top, centre; Figures 3 and 4). In contrast,
galaxies with quieter histories (e.g. bottom is yet to reignite star formation after quenching at 𝑧 = 4) show stable H i reservoirs within their half-light radius (blue
circles) that can host long-lived, stable H i discs (Figure 5).

finding at least two examples satisfying these conditions per star- 4.2 Short-lived H i discs in star-forming low-mass dwarfs
forming dwarf. As we will see in Section 4.2, these snapshots can
showcase interpretable but short-lived H i rotation curves. To quantify H i kinematics in the noisy, star-forming dwarfs, we
visually inspect individual rotation curves and H i column density
maps at the times flagged to have higher probabilities of galactic gas
Contrasting again with our star-forming examples, quiescent discs (Figure 3, diamonds). Appendix B presents the full results of
dwarfs (brown lines in Figure 3) show more stable evolution over this systematic inspection, showcasing very diverse H i distributions
time and clearer evolutionary trends. One galaxy (second row) lacks with complex spatial, kinematic and thermodynamical structures due
evidence for gas rotation (𝑣 𝜙,𝑔 /𝑣 circ ≈ 0) at all times, but the other to stellar feedback.
(fourth row; also bottom panels of Figure 2) is regularly approaching But Figure 4 (and other examples in Appendix B) show that, even
𝑣 𝜙,𝑔 /𝑣 circ ≈ 1, and 𝑣 𝜙,𝑔 /𝜎eff ≥ 1, and notably over its entire last if short-lived, organized H i rotation can occur in these systems.
billion years of evolution. As we will see in Section 4.3, this galaxy Figure 4 shows the total gas velocity profiles (left panel), surface
hosts a stable, long-lived H i disc with an easy-to-interpret rotation density and temperature profiles (middle panels) and H i column
curve. density maps viewed face-on and edge-on (right panels) of ‘Halo 600’
at 𝑡 = 11.9 Gyr. At this time, the H i distribution is spatially extended
Our analysis thus shows that intrinsic ordered gas rotation should (right panels), reaching 𝑁H i ≥ 3×1019 cm −2 outside 𝑟 1/2,3D (dotted
be expected in low-mass, H i-bearing dwarfs. However, stellar feed- and dashed lines in left panel). Such surface brightnesses are at
back in star-forming objects can efficiently disrupt small gas discs the limit of what can be achieved by deep follow-ups with current-
(𝑣 circ ≈ 10 km s −1 ), making them short-lived and rare. This pro- generation interferometers in faint dwarfs (e.g. Adams & Oosterloo
vides a natural explanation for the lack of observed rotation in the 2018). Furthermore, despite showcasing holes and being lopsided at
faintest dwarfs (e.g. Bernstein-Cooper et al. 2014; Adams & Ooster- large radii, the H i distribution is smooth and regular in the inner
loo 2018; McQuinn et al. 2021). Quiescent, H i-bearing dwarfs, that galaxy.
are yet to re-ignite star formation after cosmic reionization, offer a In fact, the 𝑣 𝜙,𝑔 profile (left, red) follows the rise of 𝑣 circ (blue)
contrastingly calmer and more stable environment. This promotes within 200 pc, as expected from equilibrium circular orbits. The
well-ordered and long-lived gas rotation, with greater prospects for H i gas also exhibits a close-to-exponential radial surface brightness
dark matter science using rotation curves which we quantify next. profile (middle, bottom), reminiscent of classical rotation curves of

MNRAS 000, 1–18 (2024)

 6 M. P. Rey et al.
Redshift Redshift
22 1 0.5 0 22 1 0.5 0
v , g /vcirc (150 pc)

(150 pc)
1 1

0 0

eff
v , g/
1 Halo 600 1
2 2
v , g /vcirc (150 pc)

Keplerian gas velocity in

(150 pc)
Keplerian gas velocity
1 equilibrium with potential 1 dominates velocity dispersion

0 0

eff
v , g/
1 Halo 600 GM: Delayed mergers 1
2 2
v , g /vcirc (150 pc)

(150 pc)
1 1

0 0
eff
v , g/

1 Halo 605 1
2 2
v , g /vcirc (150 pc)

(150 pc)

1 1

0 0
eff
v , g/

1 Halo 624 1
2 2
v , g /vcirc (150 pc)

(150 pc)

1 1

0 0
eff
v , g/

1 Halo 624 GM: Higher final mass 1

4.0 6.0 8.0 10.0 12.0 4.0 6.0 8.0 10.0 12.0
Age of the Universe (Gyr) Age of the Universe (Gyr)
Figure 3. Time evolution of 𝑣 𝜙,𝑔 /𝑣circ (left; i.e. a proxy for gas rotation in equilibrium with the gravitational potential) and 𝑣 𝜙,𝑔 /𝜎eff (right; i.e. comparing
rotational-to-pressure support) across our suite of simulated H i-bearing dwarfs after reionization. Star-forming galaxies (blue, stars marking their first time
of post-reionization star formation) show instances where they host rotationally-supported H i kinematics close to equilibrium with the gravitational potential
(𝑣 𝜙,𝑔 ≈ 𝑣circ and 𝑣 𝜙,𝑔 ≥ 𝜎eff ; marked with diamonds). These episodes are short-lived, as the H i is rapidly heated and dispersed by stellar feedback (Figure 4).
Quiescent galaxies (brown lines) have more stable kinematics, with one hosting long-lived, stable H i rotation that provides an ideal target for dark matter
inferences (Figure 5).

galactic discs. However, rotation only marginally dominates com- drift; see Appendix A for further details), we obtain the H i rotational
pared to the primary source of gas velocity dispersion (thermal pres- velocity (𝑣 rot, H i ; red, dashed) which accurately recovers 𝑣 circ deep
sure, 𝑐 𝑠 in gold) and only at specific radii. Extracting and claiming into the diffuse H i regime (𝑁H i ≥ 3 × 1019 cm −2 ; ≈ 2𝑟 1/2,3D ).
a rotational signal from moment maps will thus be challenging once These results are highly promising and show that, although rare and
observational challenges associated with such faint and small objects potentially difficult to identify, H i rotation curves can be harnessed
are folded-in (discussed further in Section 5). Nonetheless, comput- for dark matter science in star-forming low-mass dwarfs.
ing the standard pressure correction to 𝑣 𝜙,𝑔 (also called asymmetric Extending this analysis further is complicated by the unusual ther-

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 7

25
Vcirc, tot cs
104

2
v ,g

NHI 3 × 1019 cm
turb, g
Vrot, HI UVB heating
balances cooling
r1/2, 3D
20

Tg (K)
103
Velocity (km s 1 )

15 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure 4. H i kinematics in an example star-forming H i-rich low-mass dwarf galaxy (‘Halo 600’) singling a time of ordered H i rotation (𝑡 = 11.9 Gyr). The
face-on 2D tangential velocity profile of the gas (left, red) follows the rise of the 3D rotation curve (left, blue) sourced by the gravitational potential, with the H i
distribution extending to ≈ 2𝑟1/2,3D and showcasing a cold and close-to-exponential disc structure (middle panels; indicative exponential scale lengths in grey).
Stellar feedback drives asymmetric H i features in the outskirts (right panels, face-on and side-on projections in the top and bottom panels, respectively), and
photo-heating from the UV background leads to a rising thermal support (left, 𝑐𝑠 in gold and top, middle). Despite these features, traditional pressure-support
corrections to the gas velocity (dashed red; also called asymmetric drift, see Appendix A) can accurately recover 𝑣circ out to 𝑁H i ≥ 3 × 1019 cm −2 .

mal structure and density profile of the gas compared to higher- compared to more massive disc galaxies. Once viewed inclined, the
mass galaxies. The temperature is steadily rising when moving to H i linewidth from thicker discs receives contribution along the line
the outskirts (top, middle), with 𝑐 𝑠 following accordingly. Already at of sight, leading to a potential mismatch between the rotation veloc-
≈ 2 𝑟 1/2,3D , thermal pressure fully dominates rotational signals and ity measured from the H i and the intrinsic value. And even if clear
H i has transitioned from colder (𝑇 ≈ 103 K) to warmer temperatures rotation can be established, the sensitivity of these galaxies to stellar
(≈ 104 K). This transition also materializes in a change of slope of feedback makes a detailed assessment of rotation curve systematics
the gas surface density profile (middle, bottom). essential for robust dark matter inferences (see Appendix B for ex-
amples of out-of-equilibrium flows, non-circular motions, feedback-
This structure is naturally explained by the rising importance of
driven holes and Read et al. 2016b; Oman et al. 2019; Downing &
the cosmic ultraviolet background (UVB) at low-galactic masses.
Oman 2023 for further discussion).
Following cosmic reionization, the UVB provides a source of ion-
Performing these quantifications would be best undertaken by gen-
ization and heating that maintains diffuse gas in photo-ionization
erating mock H i datacubes from our simulated snapshots to as-
equilibrium around 𝑇 ≈ 104 K. Galaxies considered here have po-
sess the robustness of standards rotation curve fitting methods (e.g.
tential wells just deep enough to accrete fresh gas from their diffuse
3dbarolo; Di Teodoro & Fraternali 2015) and understand whether
surroundings (e.g. 𝑐 𝑠 is only slightly below 𝑣 circ at large radii). Gas
new approaches would be better suited to recover dark matter infor-
can self-shield and cool below 104 K in the centre of the dwarf (Rey
mation. We are currently developing a package that can easily incor-
et al. 2020) but gas in the outskirts rapidly transitions to ≈ 104 K in a
porate different pressure terms, and treat the impact of disc thickness
balance between gravity, cooling from metal lines and photo-heating
on the line-of-sight velocity distribution, and leave the quantifica-
from the UVB (e.g. Ricotti 2009; Rey et al. 2020; Benitez-Llambay
tions of these uncertainties to future work. In the next Section, we
& Frenk 2020). The detection of warm H i (104 K) in projection thus
instead focus on easier-to-interpret and long-lived H i rotation curves
cannot be unequivocally attributed to photo-heating from stars in
that can be found in quiescent dwarfs.
these faint objects, particularly when undertaking deep observations
probing the diffuse gas (e.g. Adams & Oosterloo 2018).
Establishing and interpreting H i rotation is likely to be challenging 4.3 Long-lived H i discs in quiescent dwarfs
in star-forming faint dwarfs. Although one can recover the gravita-
4.3.1 Circular, equilibrium H i rotation in a low-mass dwarf
tional potential with access to all simulated information to compute
thermal support, how to achieve this feat from H i datacubes is less Figure 5 shows the rotation curve at 𝑧 = 0 (left panel), the gas
clear. The dominance of pressure terms over rotation might point to temperature, and surface density radial and vertical profiles (right
the need to introduce new approaches to infer dark matter profiles panels) for the quiescent dwarf hosting a clear, long-lived rotation
(e.g. starting from hydrostatic equilibrium rather than axisymmetric signal (Figure 3, fourth row).
rotation; Patra 2018). Furthermore, this also yields thicker H i discs In this example, 𝑣 𝜙,𝑔 accurately tracks 𝑣 circ without corrections

MNRAS 000, 1–18 (2024)

 8 M. P. Rey et al.

25
Vcirc, tot cs

2
104

NHI 3 × 1019 cm
v ,g turb, g
UVB heating
Vrot, HI
r1/2, 3D
20 balances cooling

Tg (K)
Velocity (km s 1 )

15 103

107 Gas

zD=
10 HI

(M kpc 2 )

50 pc
rD=
100
106

pc
5

0 0.2 0.4 0.6 0.8 1.0 105 0.2 0.4 0.6 0.8 1.0 0.1 0.2 0.3 0.4 0.5
Radius (kpc) Radius (kpc) Height (kpc)
Figure 5. Classical H i disc at 𝑧 = 0 in a quiescent low-mass dwarf (‘Halo 624’). The tangential velocity profile (left, red) accurately tracks the rise of the
rotation curve (blue) within 𝑟1/2,3D (dotted). Correcting the tangential velocity for pressure support (dashed) helps modelling the rising thermal support (gold)
and recover 𝑣circ further into the outskirts of the H i distribution (dashed shows 𝑁H i ≥ 3 × 1019 cm −2 ) where the disc gets thicker (right panels). This long-lived
and easy-to-interpret H i rotation curve is unique to this object, driven by the defining shape of the host dark matter halo (Figure 6 and 7). Characterizing such a
rotation curve would prove invaluable to obtain robust inferences of inner dark matter density profiles.

to ≈ 10 km s −1 , indicating near-perfect circular rotation in equilib- in our suite does not exhibit rotation (second row in Figure 3) and
rium with the gravitational potential. Compared to our star-forming star-forming examples lack clear signals during their quiescent peri-
example (Figure 4), the inner gas is cold (right, top panels), and ods (notably before the re-ignition of their star-formation marked by
rotation strongly dominates thermal support and turbulence (𝑐 𝑠 and a star in Figure 3 ending several billion years of quiescent evolution).
𝜎turb, g in gold and brown) in the inner galaxy. Pressure corrections We thus turn next to understanding what leads to long-lived H i discs
(red, dashed) are subdominant at all radii, only becoming significant in this specific object.
when reaching more diffuse H i (≥ 𝑟 1/2,3D ; dotted line) brought to
warmer temperatures by the UVB.
The lack of disturbances from star formation in this object also en- 4.3.2 The link between H i rotation and the shape of the host dark
sures a regular, symmetric and well-ordered H i distribution (recall matter halo
Figure 2), showcasing close-to-exponential H i radial and vertical A defining feature of our quiescent galaxy with long-lived H i disc
profiles (bottom panels). The vertical profile (right panels) shows a lies in the properties of its host dark matter halo. The specific merger
thickened H i disc (aspect ratios approaching 1:2; indicative expo- history of this object, particularly a major interaction at 𝑧 ≈ 4 leads to
nential scale lengths in grey), as expected from the rising importance a strongly oblate dark matter halo shape compared to the more triax-
of pressure support towards larger radii. ial or prolate shapes across the rest of the simulated suite (𝑏/𝑎 ≈ 0.9,
To summarize, this quiescent galaxy hosts a classical H i rotation 𝑐/𝑎 ≈ 0.5 between 𝑟 ≈ 200 pc and 𝑟 ≈ 20 kpc for this halo; see
curve, at column densities achievable by current-generation radio Orkney et al. 2023, fig. 1). We link these two aspects in Figure 6,
interferometers (𝑁H i ≥ 3 × 1019 cm −2 , dashed-grey vertical line). visualizing the alignment between the H i angular momentum com-
This rotation curve is comparatively easy to interpret, holding great pared to the halo shape. We plot the H i column density map at 𝑧 = 0,
promise for extracting unbiased estimates of the inner dark matter oriented side-on compared to the angular momentum of gas with
density profiles. Even further, we show in Appendix C that a similar 𝑥H i ≥ 0.5 and overlay the 3D halo shape computed exclusively from
rotation structure is present over the last two billion years of evolution the dark matter particles as in Orkney et al. (2023) (grey mesh; whiter
of this galaxy (see also diamonds in Figure 3), with the cold and towards the foreground, blacker towards the background). Note that
circular H i rotation curve being in place for the last 500 Myr. The Orkney et al. (2023) derive halo shapes using higher-resolution re-
excellent agreement between the H i rotation and the gravitational simulations (𝑚 DM = 120 M⊙ ) of the galaxies studied in this work –
potential is thus long-lived and little disrupted. we have checked that (i) the radial profile of axis ratios at 𝑟 ≥ 100 pc
Our results strongly motivate targeting low-mass dwarfs with qui- and (ii) the presence and orientation of the gas disc at 𝑧 = 0 are
eter evolutions when searching for high-quality H i rotation curves. both consistent between the two resolutions. This also validates that
Such quiescent candidates have already been reported (e.g. Janesh the presence and formation of the H i disc is physical, rather than
et al. 2019) and their follow-up with deep and high-resolution H i stochastic or resolution-limited.
interferometers should be given high priority. However, our analysis The H i disc and the flattened axis of the oblate dark matter halo
also shows that a lack of star formation activity is insufficient to are exceptionally well aligned in Figure 6. This is best understood by
guarantee well-behaved rotation curves – the other quiescent galaxy the naturally axisymmetric geometry of a significantly oblate halo.

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 9

1021 90

2
NHI 3 × 1019 cm
80

r1/2, 3D

r200c
70
20
10 60

AM(r) AM(100pc)
NHI (cm 2 )
50
40
19
10
30

1 kpc
20
10
1018
0 1
Figure 6. H i column density map oriented edge-on with respect to the gas 10 100 101
for the galaxy hosting stable rotation (same as Figure 5). The host dark matter
halo is strongly oblate (white-grey-black mesh showcasing the 3D shape),
Radius (kpc)
defining an axisymmetric geometry well aligned with the revolution axis of
Figure 7. Orientation of the angular momentum of the gas in a radial shell
the H i disc. This configuration induces torques that align infalling gas into
compared to that in the inner 100 pc, for the same galaxy as in Figure 5. Gas
the plane of the H i disc (Figure 7) and favour its growth.
outside the virial radius (grey line) is accreted tilted compared to the inner
angular momentum (𝜃 ≤ 50◦ ), but is gradually and coherently torqued with
decreasing radius to align with the inner H i disc (𝜃 ≈ 0 for 𝑟 ≤ 300 pc). The
Such geometry induces torques that align accreting gas along its revolution axis of the H i disc also coincides with that of the dark matter halo
revolution axis, a process best studied in the case of axisymmetric shape (Figure 6).
torques induced by galactic stellar discs (see e.g. Danovich et al.
2015 for a discussion). Here, these torques are sourced by the dark
matter halo itself, as the gas and stars contribute only marginally to At this galactic mass-scale, galaxy formation effects within ΛCDM
the gravitational potential. are inefficient at dynamically heating dark matter into flat and large
To visualize this torque in action, Figure 7 shows the orientation (≈ 𝑟 1/2,3D ) dark matter cores, leading to steep dark matter den-
of the gas angular momentum in a given radial shell compared to sity profiles at 𝑟 1/2,3D in all of our dwarfs (Figure 1). Inferring the
the angular momentum of the gas in the inner 100 pc (which is al- structure of dark matter haloes in this regime, for example through
most purely H i; Figure 5). Starting from outside the virial radius H i rotation curves, thus holds great promise to pinpoint the relative
(≥ 30 kpc), gas is accreted with significant angular momentum but contributions of dark matter microphysics and galaxy formation in
orthogonal to the inner disc (𝜃 ≥ 50◦ ) before stabilizing around this driving dark matter heating.
angle across between 8 and 20 kpc. Towards smaller radii, however, We find that simulated low-mass dwarfs that are actively forming
the gas gradually gets torqued to align with the inner angular mo- stars undergo strong variability in their H i distributions, driven by
mentum of the galaxy (𝜃 ≤ 10◦ within ≈ 3𝑟 1/2,3D ), at which point the cycle of gas accretion and efficient stellar feedback (Rey et al.
it shares the same revolution axis as that of the oblate dark matter 2022). This variability is reflected in their H i kinematics, showcasing
halo shape (Figure 6). This gradual realignment of gas throughout disturbed and rapidly changing gas flows (Figure 2) as supernovae
the halo, starting at radii well outside the galaxy, firmly establishes easily disrupt gas dynamics in these shallow potential wells (𝑣 circ
the causal link between the halo shape and the presence of the H i and 𝑣 𝜙,𝑔 ≈ 10 km s −1 at 𝑟 1/2,3D ). We find occasional, short-lived
disc. (≪ 150 Myr) episodes of organized H i rotation in these star-forming
objects (Figure 3), for which rotation curves can recover the under-
lying gravitational potential (Figure 4). But the prevalence of out-of-
equilibrium feedback-driven gas flows and the (comparatively) high
5 SUMMARY AND DISCUSSION
velocity dispersions due to thermal support (𝜎eff ≈ 10 km s −1 ) lead
We have analysed the gas and H i kinematics of simulated low-mass to difficult-to-interpret rotation curves (see also Appendix B). Clear
(104 ≤ 𝑀★ ≤ 2 × 106 M⊙ ) dwarf galaxies, first introduced in Rey and robust H i rotation that can be harnessed for dark matter science
et al. (2019, 2020) and evolved to 𝑧 = 0 using high-resolution (≈ 3 pc) is thus expected to be rare in these active systems, aligning with the
zoomed cosmological simulations using the edge galaxy formation lack of observed rotation in the handful of low-mass star-forming
model (Agertz et al. 2020). We studied five dwarf galaxies that are dwarfs with detailed H i observations (e.g. Bernstein-Cooper et al.
close analogues to the observed population of faint, but gas-rich and 2014; Adams & Oosterloo 2018; McQuinn et al. 2021).
H i-bearing dwarfs (105 ≤ 𝑀H i ≤ 106 M⊙ ; see Rey et al. 2022 for a Contrastingly, two of our low-mass dwarfs undergo significantly
more detailed comparison the observed population; Irwin et al. 2007; quieter evolution, with several billion years without forming new
Cole et al. 2014; McQuinn et al. 2015, 2020, 2021; Sand et al. 2015; stars (see also Rey et al. 2020). The lack of star-formation activity
Adams & Oosterloo 2018; Brunker et al. 2019; Janesh et al. 2019; since 𝑧 ≈ 4 leads to more stable H i reservoirs in these systems with
Hargis et al. 2020; Bennet et al. 2022; Rhode et al. 2023). better organized kinematics (Figure 2 and 3). In particular, one of

MNRAS 000, 1–18 (2024)

10 M. P. Rey et al.

Halo 600 Halo 624

Inferred from mock

2
109 VLA-like data

NHI 3 × 1019 cm

NHI 3 × 1019 cm
r1/2, 3D

r1/2, 3D
(M kpc 3 )

108
Affected by numerics
DM

107

1
10 100 10 1
100
Radius (kpc) Radius (kpc)
Figure 8. Inferred dark matter density profiles (black line showing the median, grey contours the 1 and 2𝜎 intervals) from the rotation curves in Figure 4 and
Figure 5 (left and right, respectively). Both dark matter profiles are recovered within the 95% confidence intervals, although the inferred dark matter density
profile of the star-forming dwarf (left) is slightly flatter than its true profile (blue), with all other parameters (e.g. mass, concentration) poorly constrained
(Figure D1). In contrast, the dark matter density profile is well recovered all the way to the centre in the quiescent galaxy (right), ruling out a dark matter core
of size < 100 pc at 95% confidence (Figure D1). This highlights the clear prospects offered by quiescent systems, as they are more favourable to host long-lived
and organized rotation, that is easier to interpret for dark matter science.

our quiescent H i-bearing dwarf showcases a long-lived, close-to- 2012; Bonamigo et al. 2015). But their fraction is steadily rising
circular H i rotation curve (Figure 5 and Appendix C) that could be towards lower halo masses (e.g. ≈ 20 per cent of haloes with 𝑀200 =
readily and robustly interpreted for dark matter inferences. We tie the 1012 M⊙ compared to ≈ 10 for 𝑀200 = 1013 M⊙ ; Vega-Ferrero et al.
existence of this long-lived rotation curve to the specifically oblate 2017). Statistical quantifications of halo shapes across the low-mass
shape of its host dark matter halo, which plays a key role in building dwarf galaxy population remain limited in term of sample sizes (e.g.
the final H i disc by torquing circumgalactic gas to align with its Downing & Oman 2023; Orkney et al. 2023), but these estimates are
axisymmetric revolution axis (Figure 6 and 7). in line with the (very) small number statistics of one-out-five oblate
halo in our suite. Our established link between halo shapes and gas
Our results point to H i rotation being generally rare, sensitive and
rotation in small dark matter haloes makes quantitatively refining
potentially challenging to interpret in faint H i-bearing dwarfs. But
these numbers particularly pressing.
we stress that the mere existence of several examples of ordered and
easy-to-interpret H i rotation curves across a suite of only five sim-
When detected, observationally characterizing and interpreting
ulated galaxies is highly promising and strongly motivates further
H i rotation curves in such small and faint systems is likely to pose
observational and theoretical investigations. In particular, our find-
a difficult, but achievable, challenge. We highlight this point in Fig-
ings highlight clear avenues to find ‘golden eggs’ enabling robust
ure 8. To this end, we take the pressure-corrected rotation curves
dark matter inferences, that is targeting low-mass dwarfs that (i) have
presented in Figure 4 and 5 and resample them with 100 pc spa-
been quiescent for an extended period of time and have avoided rapid
tial resolution (≈ 10′′ at 2 Mpc, a spatial resolution accessible to
disruption of their gas flows by stellar feedback from newborn stars;
interferometric studies in nearby low-mass dwarfs with current in-
and (ii) are hosted in an oblate dark matter halo whose axisymmetric
struments; Bernstein-Cooper et al. 2014; Adams & Oosterloo 2018;
geometry promotes disc formation.
M. Jones et al. in preparation). Motivated by the spectral and imaging
An extended gap in star formation and a quiescent period can be sensitivity of these same studies, we then assume Gaussian velocity
inferred from a color-magnitude diagram and a lack of young, blue errors of 0.8 km s −1 and that the rotation curve is imaged down to a
stars when deep photometric imaging is available. Candidates for column density of 𝑁H i ≥ 5 × 1019 cm −2 . We then fit these mock ro-
such quiescent low-mass dwarf galaxies have in fact already been tation curves with ‘coreNFW’ dark matter density profiles using the
reported (Janesh et al. 2019, although see also Rhode et al. 2023) Bayesian approach described in Read et al. (2016b) (see also Read
but, unfortunately, the shape of their host dark matter halo cannot et al. 2017, 2019). Appendix D describes this procedure in more de-
be known a priori (or at all). Oblate dark matter halo shapes are tail and shows the posterior dark matter halo masses, concentrations
statistically rarer amongst the population of high-mass dark matter and core sizes for each dwarf galaxy. Figure 8 presents the inferred
haloes (e.g. Jing & Suto 2002; Macciò et al. 2007; Schneider et al. dark matter density profiles, with their median and 1-2𝜎 intervals

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 11
(black line and grey contours) compared to the ‘true’ simulated pro- A key aspect to achieving this goal will be to obtain robust pre-
files (coloured contours, with shading showing 1-2𝜎 uncertainties dictions of the connection between dark matter and H i properties in
from the Poisson noise in each radial bin). low-mass dwarf galaxies. This requires us to pinpoint the coupling
For both the star-forming galaxy (left) and the quiescent galaxy between a dwarf’s ISM and stellar feedback, which is key not only to
(right), the inferred dark matter profiles are compatible with their un- regulate the ability of supernova-driven outflows to drive dark matter
derlying simulated ‘truth’ at 2𝜎. In both cases, this is mainly driven by heating, but also their ability to disrupt H i discs. Many galaxy forma-
an underestimated normalization of the density profile (i.e. the overall tion models, including our own, now converge in predicting that the
dark matter halo mass). This entirely reflects the limited radial range low stellar masses of faint dwarfs does not provide enough SN energy
of realistic rotation curves, as current studies in this regime could to fully heat their central dark matter into a large and flat dark matter
only image the rising part of the rotation curve, leading to largely core (e.g. Peñarrubia et al. 2012; Di Cintio et al. 2014; Chan et al.
unconstrained masses and concentrations (Appendix D). We verified 2015; Oñorbe et al. 2015; Tollet et al. 2016; Lazar et al. 2020; Orkney
that halo masses are much more accurately and precisely recovered et al. 2021). This study provides the first link between these results
if providing the rotation curve down to 𝑁H i ≥ 5 × 1017 cm −2 . and the efficiency of H i disc formation at this galactic mass-scale.
In the star-forming case, the inferred profile is flatter than the Despite these achievements and the accurate modelling of supernova
truth at 𝑟 ≤ 𝑟 1/2,3D , with the inference showing marginal preference explosions in our simulations, further quantifications are required to
for a non-zero core of size ≈ 0.5 𝑟 1/2,3D (Figure D1). This reflects better understand the robustness of our predicted H i kinematics. In
the challenges of modelling these systems, as feedback-driven non- particular, photo-ionization feedback can lead to a more gentle and
circular motions or out-of-equilibrium bulk flows (such as those less explosive regulation of star formation (e.g. Agertz et al. 2020;
visible in Figure 4 that lead to the dip in 𝑣 rot, H i at ≈ 200 pc) can bias Smith et al. 2021) and could further promote H i disc formation.
rotation curves inferences towards flatter profiles (see e.g. discussion Resimulating all of our dwarfs accounting for radiative effects and
in Oman et al. 2019). These effects are further compounded by the improved tracking of gas flows over time (Cadiou et al. 2019) will
limited radial range of the rotation curve, and thus their constraining be tackled in future work (Rey et al. in preparation), allowing us to
power on other halo parameters. pinpoint how gas spirals into and flows out of these sensitive objects.
Nonetheless, in the case of the quiescent galaxy with a well-ordered
rotation curve, the inference confidently recovers a steep dark matter
profile all the way to small radii, ruling out a dark matter core of size <
100 pc at 95% confidence (Figure D1). If confirmed observationally,
ACKNOWLEDGEMENTS
such steep profile would establish that SN-driven dark matter heating
cease to be efficient in small galaxies. This, in turn, would put a strong MR would like to thank Betsey Adams, Erwin de Blok, Corentin
and clean bound on dark matter models that flatten density profiles Cadiou and Filippo Fraternali for insightful discussions during the
in dwarf galaxies (e.g. self-interactions, axions). construction of this work and comments on earlier versions of this
This proof-of-concept inference is far from a complete end-to-end manuscript. We would like to thank the anonymous referee for a con-
validation and makes significant simplifying assumptions (e.g. pres- structive review that improved the quality of the manuscript. MR is
sure corrections are derived from simulated profiles rather than a H i supported by the Beecroft Fellowship funded by Adrian Beecroft.
datacube, galaxy inclination is assumed to be perfectly known; see MO acknowledges the UKRI Science and Technology Facilities
Read et al. 2017; Oman et al. 2019; McQuinn et al. 2021; Downing Council (STFC) for support (grant number ST/R505134/1). OA ac-
& Oman 2023 for further discussion of these challenges). Nonethe- knowledges support from the Knut and Alice Wallenberg Founda-
less, it demonstrates the incredible potential of targeting low-mass tion, the Swedish Research Council (grant number 2019-04659), the
galaxies that have been quiescent for a while. These systems not only Royal Physiographic Society of Lund and the Swedish National Space
are more favourable to host long-lived and organized H i rotation, Agency (SNSA Dnr2023-00164). AP is supported by the Royal Soci-
but their rotation curves are also easier to interpret as star formation ety. AAP acknowledges support of the STFC consolidated grant num-
and supernovae have had less opportunity to dynamically heat dark bers [ST/S000488/1] and [ST/W000903/1]. WM thanks the Science
matter. Finding these ‘golden egg’ systems will thus be key to unlock and Technology Facilities Council (STFC) Centre for Doctoral Train-
high-quality, precise and unbiased dark matter inferences. ing (CDT) in Data intensive Science at the University of Cambridge
Excitingly, ongoing wide-sky H i surveys such as Wallaby on (STFC grant number 2742968) for a PhD studentship. This project
the Australian Square Kilometer Array Pathfinder (Koribalski et al. has received funding from the European Union’s Horizon 2020 re-
2020), Apertif-Medium deep on the Westerbrook Synthesis Radio search and innovation programme under grant agreement number
Telescope (van Cappellen et al. 2022) and efforts with the Five- 818085 GMGalaxies. This work was performed using the DiRAC
hundred-meter Aperture Spherical radio Telescope (Kang et al. 2022) Data Intensive service at Leicester, operated by the University of Le-
will soon provide many new candidates for H i-rich, low-mass dwarfs. icester IT Services, which forms part of the STFC DiRAC HPC Fa-
At the same time, serendipitous detections of such objects are becom- cility (www.dirac.ac.uk). The equipment was funded by BEIS capital
ing more common (e.g. Brunker et al. 2019; Hargis et al. 2020; Ben- funding via STFC capital grants ST/K000373/1 and ST/R002363/1
net et al. 2022; Jones et al. 2023), particularly in the realm of sensitive and STFC DiRAC Operations grant ST/R001014/1. DiRAC is part
H i surveys cross-matched with deep imaging like MHONGOOSE of the National e-Infrastructure. The authors acknowledge the use of
and MIGHTEE on MeerKat (de Blok et al. 2020; Maddox et al. the UCL Grace High Performance Computing Facility, the Surrey
2021). These new catalogues and detections will lack the necessary Eureka supercomputer facility, and their associated support services.
combination of angular and spectral resolution to obtain robust H i This work was partially supported by the UCL Cosmoparticle Initia-
kinematics, but they will be invaluable to select the most promising tive.
targets for high-quality rotation curve follow-ups with deep interfer- We thank the developers and maintainers of pynbody (Pontzen
ometry from VLA and MeerKAT. Combined, these new capabilities et al. 2013), tangos (Pontzen & Tremmel 2018), numpy (van der
will ensure we meet the clear promises offered by the very faint end Walt et al. 2011), scipy (Virtanen et al. 2020), jupyter (Ragan-
of the H i-bearing population to constrain dark matter physics. Kelley et al. 2014), matplotlib (Hunter 2007), the Astrophysics

MNRAS 000, 1–18 (2024)

12 M. P. Rey et al.
Data Service and the arXiv preprint repository for providing open- Dalcanton J. J., Stilp A. M., 2010, ApJ, 721, 547
source softwares and services that were used extensively in this work. Danovich M., Dekel A., Hahn O., Ceverino D., Primack J., 2015, MNRAS,
The main roles of the authors were, using the CRediT (Contribu- 449, 2087
tion Roles Taxonomy) system1 : De Leo M., Read J. I., Noel N. E. D., Erkal D., Massana P., Carrera R., 2023,
MR: Conceptualization ; Data curation; Formal analysis; Investi- Surviving the Waves: Evidence for a Dark Matter Cusp in the Tidally
Disrupting Small Magellanic Cloud (arxiv:2303.08838)
gation; Writing – original draft. MO: Data Curation; Formal analysis;
Di Cintio A., Brook C. B., Macciò A. V., Stinson G. S., Knebe A., Dutton
Writing – review and editing. JR: Conceptualization; Formal analy- A. A., Wadsley J., 2014, MNRAS, 437, 415
sis; Resources; Writing – review and editing. PD: Conceptualization; Di Teodoro E. M., Fraternali F., 2015, MNRAS, 451, 3021
Writing – review and editing. OA: Methodology; Software; Writing – Downing E. R., Oman K. A., 2023, MNRAS, 522, 3318
review and editing. AP: Writing – review and editing. AAP: Writing Efstathiou G., 1992, MNRAS, 256, 43P
– review and editing. SK: Conceptualization. WM: Writing – review Eisenstein D. J., Hut P., 1998, ApJ, 498, 137
and editing. El-Zant A., Shlosman I., Hoffman Y., 2001, ApJ, 560, 636
Enzi W., et al., 2021, MNRAS, 506, 5848
Fitts A., et al., 2017, MNRAS, 471, 3547
Flores R. A., Primack J. R., 1994, ApJ, 427, L1
DATA AVAILABILITY Foreman-Mackey D., Hogg D. W., Lang D., Goodman J., 2013, PASA, 125,
306
The data underlying this article will be shared on reasonable request
Garrison-Kimmel S., et al., 2017, MNRAS, 471, 1709
to the corresponding author. Geha M., Blanton M. R., Yan R., Tinker J. L., 2012, ApJ, 757, 85
Gilman D., Birrer S., Nierenberg A., Treu T., Du X., Benson A., 2020,
MNRAS, 491, 6077
Girardi L., et al., 2010, ApJ, 724, 1030
REFERENCES
Gnedin N. Y., 2000, ApJ, 542, 535
Adams E. A. K., Oosterloo T. A., 2018, A&A, 612, A26 Goerdt T., Moore B., Read J. I., Stadel J., 2010, ApJ, 725, 1707
Agertz O., et al., 2020, MNRAS, 491, 1656 Haardt F., Madau P., 1996, ApJ, 461, 20
Alam S., et al., 2021, PRD, 103, 083533 Hargis J. R., et al., 2020, ApJ, 888, 31
Amorisco N. C., 2017, ApJ, 844, 64 Hoeft M., Yepes G., Gottlöber S., Springel V., 2006, MNRAS, 371, 401
Anderhalden D., Schneider A., Macciò A. V., Diemand J., Bertone G., 2013, Hsueh J. W., Enzi W., Vegetti S., Auger M. W., Fassnacht C. D., Despali G.,
JCAP, 2013, 014 Koopmans L. V. E., McKean J. P., 2020, MNRAS, 492, 3047
Armengaud E., Palanque-Delabrouille N., Yèche C., Marsh D. J. E., Baur J., Hunter J. D., 2007, CiSE, 9, 90
2017, MNRAS, 471, 4606 Iorio G., Fraternali F., Nipoti C., Di Teodoro E., Read J. I., Battaglia G., 2017,
Aubert D., Teyssier R., 2010, ApJ, 724, 244 ApJ, 466, 4159
Banik N., Bovy J., Bertone G., Erkal D., de Boer T. J. L., 2021, JCAP, 2021, Iršič V., Viel M., Haehnelt M. G., Bolton J. S., Becker G. D., 2017, PRL, 119,
043 031302
Benitez-Llambay A., Frenk C., 2020, MNRAS, 498, 4887 Irwin M. J., et al., 2007, ApJ, 656, L13
Benitez-Llambay A., Fumagalli M., 2021, ApJ, 921, L9 Janesh W., Rhode K. L., Salzer J. J., Janowiecki S., Adams E. A. K., Haynes
Benítez-Llambay A., Navarro J. F., Abadi M. G., Gottlöber S., Yepes G., M. P., Giovanelli R., Cannon J. M., 2019, AJ, 157, 183
Hoffman Y., Steinmetz M., 2015, MNRAS, 450, 4207 Jeon M., Besla G., Bromm V., 2017, ApJ, 848, 85
Bennet P., et al., 2022, ApJ, 924, 98 Jethwa P., Erkal D., Belokurov V., 2018, MNRAS, 473, 2060
Bernstein-Cooper E. Z., et al., 2014, AJ, 148, 35 Jing Y. P., Suto Y., 2002, ApJ, 574, 538
Bertone G., Tait T. M. P., 2018, Nature, 562, 51 Jones M. G., et al., 2023, ApJL, 957, L5
Bertone G., Hooper D., Silk J., 2005, Physics Reports, 405, 279 Kang J., Zhu M., Ai M., Yu H., Sun C., 2022, RAA, 22, 065019
Binney J., Tremaine S., 2008, Galactic Dynamics: Second Edition Kennedy R., Frenk C., Cole S., Benson A., 2014, MNRAS, 442, 2487
Boehm C., Schewtschenko J. A., Wilkinson R. J., Baugh C. M., Pascoli S., Kim S. Y., Peter A. H. G., Hargis J. R., 2018, PRL, 121, 211302
2014, MNRAS, 445, L31 Klypin A., Gottlöber S., Kravtsov A. V., Khokhlov A. M., 1999, ApJ, 516,
Bonamigo M., Despali G., Limousin M., Angulo R., Giocoli C., Soucail G., 530
2015, MNRAS, 449, 3171 Koribalski B. S., et al., 2020, ASS, 365, 118
Bouché N. F., et al., 2022, A&A, 658, A76 Lazar A., et al., 2020, MNRAS, 497, 2393
Brunker S. W., et al., 2019, AJ, 157, 76 Ledinauskas E., Zubovas K., 2018, A&A, 615, A64
Bullock J. S., Boylan-Kolchin M., 2017, ARA&A, 55, 343 Lelli F., McGaugh S. S., Schombert J. M., 2016, AJ, 152, 157
Bureau M., Carignan C., 2002, AJ, 123, 1316 Macciò A. V., Dutton A. A., van den Bosch F. C., Moore B., Potter D., Stadel
Burkert A., 2000, ApJ, 534, L143 J., 2007, MNRAS, 378, 55
Cadiou C., Dubois Y., Pichon C., 2019, A&A, 621, A96 Maddox N., et al., 2021, A&A, 646, A35
Chan T. K., Kereš D., Oñorbe J., Hopkins P. F., Muratov A. L., Faucher- Malhan K., Valluri M., Freese K., Ibata R. A., 2022, ApJL, 941, L38
Giguère C. A., Quataert E., 2015, MNRAS, 454, 2981 Mancera Piña P. E., et al., 2020, MNRAS, 495, 3636
Clementini G., Cignoni M., Contreras Ramos R., Federici L., Ripepi V., McQuinn K. B. W., et al., 2015, ApJ, 812, 158
Marconi M., Tosi M., Musella I., 2012, ApJ, 756, 108 McQuinn Kristen. B. W., et al., 2020, ApJ, 891, 181
Clowe D., Bradač M., Gonzalez A. H., Markevitch M., Randall S. W., Jones McQuinn K. B. W., et al., 2021, ApJ, 918, 23
C., Zaritsky D., 2006, ApJ, 648, L109 Moore B., 1994, Nature, 370, 629
Cole A. A., Weisz D. R., Dolphin A. E., Skillman E. D., McConnachie A. W., Moore B., Ghigna S., Governato F., Lake G., Quinn T., Stadel J., Tozzi P.,
Brooks A. M., Leaman R., 2014, ApJ, 795, 54 1999, ApJ, 524, L19
Collins M. L. M., Read J. I., 2022, Nat Ast, 6, 647 Naab T., Ostriker J. P., 2017, ARA&A, 55, 59
Courty S., Alimi J. M., 2004, A&A, 416, 875 Nadler E. O., et al., 2020, ApJ, 893, 48
Nadler E. O., et al., 2021a, PRL, 126, 091101
Nadler E. O., Birrer S., Gilman D., Wechsler R. H., Du X., Benson A.,
1 https://authorservices.wiley.com/author-resources/ Nierenberg A. M., Treu T., 2021b, ApJ, 917, 7
Journal-Authors/open-access/credit.html Navarro J. F., Eke V. R., Frenk C. S., 1996, MNRAS, 283, L72

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 13
Nipoti C., Binney J., 2015, MNRAS, 446, 1820 Sawala T., et al., 2016, MNRAS, 457, 1931
Noh Y., McQuinn M., 2014, MNRAS, 444, 503 Schive H.-Y., Liao M.-H., Woo T.-P., Wong S.-K., Chiueh T., Broadhurst T.,
Nori M., Baldi M., 2021, MNRAS, 501, 1539 Hwang W. Y. P., 2014, PRL, 113, 261302
Oh S.-H., de Blok W. J. G., Brinks E., Walter F., Kennicutt Jr. R. C., 2011, Schneider M. D., Frenk C. S., Cole S., 2012, JCAP, 2012, 030
AJ, 141, 193 Schumann M., 2019, Journal of Physics G Nuclear Physics, 46, 103003
Oh S.-H., et al., 2015, AJ, 149, 96 Smith M. C., Bryan G. L., Somerville R. S., Hu C.-Y., Teyssier R., Burkhart
Okamoto T., Gao L., Theuns T., 2008, MNRAS, 390, 920 B., Hernquist L., 2021, MNRAS, 506, 3882
Oman K. A., Marasco A., Navarro J. F., Frenk C. S., Schaye J., Benítez- Somerville R. S., 2002, ApJ, 572, L23
Llambay A., 2019, MNRAS, 482, 821 Somerville R. S., Davé R., 2015, ARA&A, 53, 51
Oñorbe J., Boylan-Kolchin M., Bullock J. S., Hopkins P. F., Kereš D., Faucher- Spergel D. N., Steinhardt P. J., 2000, PRL, 84, 3760
Giguère C.-A., Quataert E., Murray N., 2015, MNRAS, 454, 2092 Stopyra S., Pontzen A., Peiris H., Roth N., Rey M. P., 2021, ApJS, 252, 28
Orkney M. D. A., et al., 2021, MNRAS, 504, 3509 Teyssier R., 2002, A&A, 385, 337
Orkney M. D. A., et al., 2022, MNRAS, 515, 185 Teyssier R., Pontzen A., Dubois Y., Read J. I., 2013, MNRAS, 429, 3068
Orkney M. D. A., Taylor E., Read J. I., Rey M. P., Pontzen A., Agertz O., Tollet E., et al., 2016, MNRAS, 456, 3542
Kim S. Y., Delorme M., 2023, MNRAS, 525, 3516 Valenzuela O., Rhee G., Klypin A., Governato F., Stinson G., Quinn T.,
Ott J., et al., 2012, AJ, 144, 123 Wadsley J., 2007, ApJ, 657, 773
Patra N. N., 2018, MNRAS, 480, 4369 Vega-Ferrero J., Yepes G., Gottlöber S., 2017, MNRAS, 467, 3226
Peñarrubia J., Pontzen A., Walker M. G., Koposov S. E., 2012, ApJ, 759, L42 Veltmaat J., Niemeyer J. C., Schwabe B., 2018, PRD, 98, 043509
Pereira Wilson M., Navarro J., Santos Santos I., Benitez Llambay A., 2023, Verbeke R., Papastergis E., Ponomareva A. A., Rathi S., De Rijcke S., 2017,
MNRAS, 519, 1425 A&A, 607, A13
Pineda J. C. B., Hayward C. C., Springel V., de Oliveira C. M., 2017, MNRAS, Virtanen P., et al., 2020, Nat Methods, 17, 261
466, 63 Vogelsberger M., Zavala J., Schutz K., Slatyer T. R., 2019, MNRAS, 484,
Planck Collaboration et al., 2020, A&A, 641, A6 5437
Polisensky E., Ricotti M., 2011, PRD, 83, 043506 Weisz D. R., et al., 2012, ApJ, 748, 88
Pontzen A., Governato F., 2012, MNRAS, 421, 3464 Yasin T., Desmond H., Devriendt J., Slyz A., 2023, MNRAS, p. stad1183
Pontzen A., Governato F., 2014, Nature, 506, 171 Zwicky F., 1933, Helvetica Physica Acta, 6, 110
Pontzen A., Tremmel M., 2018, ApJS, 237, 23 de Blok W. J. G., Bosma A., 2002, A&A, 385, 816
Pontzen A., Roškar R., Stinson G., Woods R., 2013, Astrophysics Source de Blok W. J. G., et al., 2020, A&A, 643, A147
Code Library, p. ascl:1305.002 van Albada T. S., Bahcall J. N., Begeman K., Sancisi R., 1985, ApJ, 295, 305
Pontzen A., Rey M. P., Cadiou C., Agertz O., Teyssier R., Read J., Orkney M. van Cappellen W. A., et al., 2022, A&A, 658, A146
D. A., 2021, MNRAS, 501, 1755 van der Walt S., Colbert S. C., Varoquaux G., 2011, CiSE, 13, 22
Posti L., Fraternali F., Marasco A., 2019, A&A, 626, A56
Power C., Navarro J. F., Jenkins A., Frenk C. S., White S. D. M., Springel V.,
Stadel J., Quinn T., 2003, MNRAS, 338, 14
APPENDIX A: PRESSURE SUPPORT AND DERIVATIONS
Putman M. E., Zheng Y., Price-Whelan A. M., Grcevich J., Johnson A. C.,
Tollerud E., Peek J. E. G., 2021, ApJ, 913, 53 In this Appendix, we revisit the formal framework behind rotation
Ragan-Kelley M., Perez F., Granger B., Kluyver T., Ivanov P., Frederic J., curve analysis to evaluate the validity of common analysis assump-
Bussonnier M., 2014, American Geophysical Union, 2014, H44D tions in the extreme galactic regime that we are considering.
Read J. I., Erkal D., 2019, MNRAS, 487, 5799
We start from the radial component of the Euler equation in cylin-
Read J. I., Gilmore G., 2005, MNRAS, 356, 107
Read J. I., Wilkinson M. I., Evans N. W., Gilmore G., Kleyna J. T., 2006,
drical coordinates
MNRAS, 367, 387 2
𝜕𝑣 𝑅,𝑔 𝜕𝑣 𝑅,𝑔 𝑣 𝜙,𝑔 1 𝜕𝑃 𝜕Φ
Read J. I., Agertz O., Collins M. L. M., 2016a, MNRAS, 459, 2573 + 𝑣 𝑅,𝑔 − =− − , (A1)
𝜕𝑡 𝜕𝑅 𝑅 𝜌 𝜕𝑅 𝜕𝑅
Read J. I., Iorio G., Agertz O., Fraternali F., 2016b, MNRAS, 462, 3628
Read J. I., Iorio G., Agertz O., Fraternali F., 2017, MNRAS, 467, 2019 where 𝑣 𝑅,𝑔 and 𝑣 𝜙,𝑔 are the gas radial and azimuthal velocities, 𝑃 is
Read J. I., Walker M. G., Steger P., 2019, MNRAS, 484, 1401 the gas pressure, Φ is the gravitational potential and 𝑅 the cylindrical
Rey M. P., Pontzen A., Agertz O., Orkney M. D. A., Read J. I., Saintonge A., radius.
Pedersen C., 2019, ApJL, 886, L3 Assuming that the system is in equilibrium (𝜕𝑣 𝑅,𝑔 /𝜕𝑡 = 0), that
Rey M. P., Pontzen A., Agertz O., Orkney M. D. A., Read J. I., Rosdahl J., radial non-circular motions are negligible (𝑣 𝑅,𝑔 𝜕𝑣 𝑅,𝑔 /𝜕𝑅 ≈ 0) and
2020, MNRAS, 497, 1508
Rey M. P., Pontzen A., Agertz O., Orkney M. D. A., Read J. I., Saintonge A.,
defining 𝑣 2circ = 𝑅 𝜕Φ/𝜕𝑅, we obtain
Kim S. Y., Das P., 2022, MNRAS, 511, 5672 𝑅 𝜕𝑃
Rhode K. L., et al., 2013, AJ, 145, 149 𝑣 2circ = 𝑣 2𝜙,𝑔 − . (A2)
𝜌 𝜕𝑅
Rhode K. L., Smith N. J., Janesh W. F., Salzer J. J., Adams E. A. K., Haynes
M. P., Janowiecki S., Cannon J. M., 2023, AJ, 166, 113 Here, 𝑣 circ is sourced by all gravitating components (dark matter, stars
Ricotti M., 2009, MNRAS, 392, L45 and gas) and can be inferred from 𝑣 𝜙,𝑔 by inverting Equation A2. In
Rogers K. K., Peiris H. V., 2021, PRL, 126, 071302 the absence of pressure gradients throughout the disc, Equation A2,
Romano-Díaz E., Shlosman I., Heller C., Hoffman Y., 2009, ApJ, 702, 1250 reduces to Keplerian circular motions 𝑣 2circ = 𝑣 2𝜙,𝑔 and the inference
Rosdahl J., Blaizot J., 2012, MNRAS, 423, 344 is trivial.
Rubin V. C., Ford Jr. W. K., 1970, ApJ, 159, 379 But more generally, the measured rotational velocity 𝑣 𝜙,𝑔 needs to
Rubin V. C., Ford Jr. W. K., Thonnard N., 1980, ApJ, 238, 471
be corrected to account for the additional pressure support. This term
Sales L. V., Wetzel A., Fattahi A., 2022, Nat Ast, 6, 897
is often called an asymmetric drift correction due to its similarity with
Sand D. J., Spekkens K., Crnojević D., Hargis J. R., Willman B., Strader J.,
Grillmair C. J., 2015, ApJL, 812, L13 the asymmetric drift derived from Jeans’ equations of collisionless
Sanders J. L., Evans N. W., Dehnen W., 2018, MNRAS, 478, 3879 dynamics (see Binney & Tremaine 2008, subsection 4.8 and Pineda
Sardone A., Peter A. H. G., Brooks A. M., Kaczmarek J., 2023, Closing et al. 2017 for further discussion of the formally distinct assumptions
the Gap between Observed Low-Mass Galaxy HI Kinematics and CDM underlying these two derivations).
Predictions (arxiv:2306.07417) Since 𝑃 is not directly observable, we need further assumptions

MNRAS 000, 1–18 (2024)

14 M. P. Rey et al.
to observationally estimate pressure support across the disc. Further APPENDIX B: ADDITIONAL H i ROTATION CURVES FOR
assuming that the gas is ideal and that its pressure is dominated by STAR-FORMING DWARFS
isotropic processes (i.e. thermal and kinetic turbulence rather than
In this Appendix, we provide additional visualizations of the rotation
possibly anisotropic non-thermal processes such as magnetic fields
2 = 𝑐2 + 𝜎2 curves and H i column density maps for star-forming dwarfs at the
or cosmic rays), we can write 𝑃 = 𝜌 𝜎eff where 𝜎eff 𝑠 turb, g
.
√︁ times of higher probability of gas rotation identified in Section 4
Here 𝑐 𝑠 = 𝑃/𝜌 is the isothermal sound speed (applicable for our (marked as diamonds in Figure 3). All velocity profiles, rotation
√︃
regime of fast thermal equilibrium) and 𝜎turb, g = 𝜎𝑥2 + 𝜎𝑦2 + 𝜎𝑧2 curves, radial profiles and H i column density maps are obtained as
is the isotropic 3D gas velocity dispersion estimated in each gas cell described in Section 3 and match the presentation of Figure 4. The
from the velocities of its 8-closest AMR neighbours2 . following figures show each individual flagged snapshots of star-
forming dwarfs and Appendix C presents further examples of our
Injecting these terms in Equation A2 then leads to
long-lived rotation curve.
Figures B1, B2 and B3 show three additional timestamps of the
same galaxy as in Figure 4 (‘Halo 600’), further demonstrating the
2 time variability and complexity of H i kinematics driven by efficient
𝑅 𝜕 (𝜌𝜎eff )
𝑣 2𝑐 = 𝑣 2𝜙 − ≡ 𝑣 rot,𝑔 , (A3) stellar feedback in shallow potential wells:
𝜌 𝜕𝑅
• Figure B1 shows a compact and strongly asymmetric H i morphol-
ogy, where the dominating rotation signal is difficult to interpret due
to the multimodal and offset H i distribution (see also Figure B7 for
where 𝑣 rot,𝑔 is the effective rotational velocity. another example in a different object).
Observations can only measure 2D surface densities, Σ𝑔 , rather • Figure B2 shows an extended and visually flattened H i distribution
than the 3D gas density 𝜌. But assuming that the disc structure with a 𝑣 𝜙,𝑔 rotation profile (left, red) matching 𝑣 circ (blue) that
decouples in the radial and vertical directions (e.g. exponential in 𝑧 dominates the velocity dispersion terms (brown and gold). Together
with a scale-height independent of radius), we can obtain with Figures 4 and B8 in a different object, these are our clearer cases
of H i rotation.
• Figure B3 shows an extended H i distribution strongly affected by
2
stellar feedback, with high velocity dispersions and 𝑣 rot,𝑔 that do not
𝑅 𝜕 (Σ𝑔 (𝑅)𝜎eff (𝑅)) recover 𝑣 circ (see also Figures B4, B5 and B6).
𝑣 2𝑐 = 𝑣 2𝜙 − . (A4)
Σ𝑔 (𝑅) 𝜕𝑅
To summarize, our systematic investigation successfully flags
short-lived H i rotation curves in star-forming low-mass dwarfs. Even
a single example of an easy-to-interpret rotation curve would prove
Taking Σ𝑔 = ΣH i in Equation A4 recovers the commonly used powerful for dark matter inferences, but we stress that further work is
correction to derive 𝑣 rot, H i accounting for pressure support (e.g. Bu- needed to ensure robust inferences. In particular, out-of-equilibrium
reau & Carignan 2002; Valenzuela et al. 2007; Oh et al. 2011, 2015; and non-circular motions driven by stellar feedback seem prevalent
Dalcanton & Stilp 2010; Read et al. 2016b; Iorio et al. 2017; Pineda in our star-forming low-mass dwarfs and are known to bias inferences
et al. 2017; Oman et al. 2019), where all terms are observable and can of dark matter properties (e.g. Read et al. 2016b; Oman et al. 2019;
be derived from H i datacubes and their moment 0, 1, and 2 integrals. Downing & Oman 2023).
Due to its 1/ΣH i (𝑅) scaling, the pressure contribution is expected to
be small at higher densities and inner radii, but become increasingly
important towards larger radii. Formally, −𝜕 (ΣH i 𝜎eff 2 )/𝜕𝑅 can take
APPENDIX C: EARLIER EXAMPLES OF THE
any sign depending on the radial profiles at hand. But for typical LONG-LIVED H i ROTATION CURVE
galactic applications, ΣH i (𝑅) exponentially declines while 𝜎eff (𝑅)
is more slowly varying across the cold/warm ISM traced by H i, In this Appendix, we demonstrate that the rotation curve presented
leading to a positive contribution to 𝑣 𝜙 in Equation A4. in Figure 5 at 𝑧 = 0 is long-lived. Figures C1 and C2 show the
velocity profiles, surface density profiles and H i column density
In the analysis of this study, we determine 2D radial profiles in
maps at 𝑡 = 11.9 Gyr and 13.4 Gyr. In both cases, we recover the
a thin slab at 𝑧 = 0 height of 𝑣 𝜙,𝑔 , 𝑐 𝑠 , 𝜎turb, g , Σ𝑔 and smooth
flattened, close-to-exponential H i distributions with a clear rotation
them with a 10-pixel Gaussian filter to avoid noise when estimating
signal that accurately recover 𝑣 circ once pressure support is accounted
gradients numerically. We then apply Equation A4 to determine the
for (𝑣 rot, H i ).
pressure-corrected 𝑣 rot,𝑔 shown as a red dotted line in Figure 4 and 5
This rotation curve is thus in place over several billion years, mak-
and across Appendix A.
ing it a prime target for dark matter science. Gas can still be disrupted
As discussed in Section 4, this traditional correction is adequate by stellar feedback from old stellar populations (e.g. SNIa and AGB
and improves the recovery 𝑣 circ from 𝑣 𝜙,𝑔 within 𝑟 1/2,3D when the stars; see Rey et al. 2020 for further discussion), with the gap in rota-
H i distribution is stable over time and well organized (Figure 5). tion around 12.0 Gyr being due to a single SNIa explosion. But these
When active star formation and associated stellar feedback dominate events become intrinsically rarer towards 𝑧 = 0 and, as expected from
the dynamics of the gas, however, this is much less clear. Assumptions the torques exerted by the host dark matter halo (Section 4.3.2), the
of steady-state equilibrium and circular motions are likely violated by gas disc rapidly reforms to host stable H i rotation.
feedback-driven bulk flows, while 𝑐 𝑠 profiles rising in the outskirts
due to the UV background can lead to both positive and negative
pressure corrections. This points to a need for new methods to recover 2 We use a spline Kernel as implemented by the pynbody library to weight the
dark matter information from star-forming faint dwarfs which we will velocity dispersion, and verified that using an unweighted velocity dispersion,
develop in forthcoming work. or using the 64 closest AMR neighbours does not impact our conclusions

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 15

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B1. Same object as in Figure 4 at 𝑡 = 11.1 Gyr. The H i distribution is strongly asymmetric, showing a potential small, cold, and rotating H i disc within
𝑟1/2,3D , but lacking ordered rotation on larger scales.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20
Tg (K)

103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI
(M kpc 2 )

rD=
100

106
pc

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B2. Same object as in Figure 4 at 𝑡 = 13.1 Gyr. The H i distribution is extended and visually flattened on scales larger than 𝑟1/2,3D , with a tangential
velocity curve close to the gravitational potential. But the offset between the H i distribution and the bottom of the potential well (top, right) would likely
complicate a dark matter inference.

APPENDIX D: DARK MATTER DENSITY PROFILE 5 corrected for pressure support (𝑣 rot, H i , dashed red). We resample
INFERENCE these rotation curves with a spatial sampling of 100 pc (i.e. 10”
at 2 Mpc), assign Gaussian errors of 0.8 km s −1 to each velocity
In this Appendix, we describe the setup used to infer the dark matter point and truncate the rotation curve at 𝑁H i ≥ 5 × 1019 cm −2 . All
density profiles presented in Figure 8. these numbers are motivated by the sensitivity, velocity and spatial
We construct mock data from the rotation curves of Figure 4 and resolution currently achievable by deep studies with for example the

MNRAS 000, 1–18 (2024)

 16 M. P. Rey et al.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B3. Same object as in Figure 4 but at 13.2 Gyr. The H i distribution is extended (out to 3 𝑟1/2,3D ) but strongly affected by stellar feedback events that
drive complex spatial structure and irregular rotation and temperature profiles.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20
Tg (K)

103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI
(M kpc 2 )

rD=
100

106
pc

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B4. Same as Figure 4 but showing ‘Halo 605’ at 11.5 Gyr. The H i distribution shows signs of rotation (left) that has just been disrupted by a feedback
event, driving a large H i hole in the centre (top, right).

VLA or MeerKAT (e.g. Bernstein-Cooper et al. 2014; Adams & matter flattening in units of 𝑟 1/2,3D , and 𝜅 encoding the slope of this
Oosterloo 2018). flattening. All analytics and details about the ‘coreNFW’ profile are
We fit these mock rotation curve with ‘coreNFW’ profiles using described in Read et al. (2016a).
the inference framework of Read et al. (2016b) (see also Read et al. The Bayesian inference is performed using Markov-Chain Monte
2017, 2019 for similar inferences). The ‘coreNFW’ profile has four Carlo from the emcee package (Foreman-Mackey et al. 2013). We run
parameters: the NFW virial mass 𝑀200 and halo concentration 𝑐 200𝑐 , the chain until convergence (50 autocorrelation times) and discard
the core parameter 𝜂 that encodes a characteristic size of the dark the first 10% samples as burn in. We assume wide and flat priors on

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 17

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B5. Same object as in Figure B4 but at 12.3 Gyr. The H i distribution is visually extended and irregular (right panels) with a clearly peaking 𝑣 𝜙,𝑔 profile
(left). But this signal is dominated by the thermal support of a warm H i (top, middle), making the interpretation of 𝑣rot, H i challenging.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20
Tg (K)

103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI
(M kpc 2 )

rD=
100

106
pc

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B6. Same object as in Figure B4 but at 12.9 Gyr. Again, the H i distribution is visually extended and flattened (right panels) but the thermal support of
the warm H i dominates over the rotational signal (left).

𝑀200 and 𝑐 200𝑐 (108 ≤ 𝑀200 ≤ 1010 M⊙ , 5 ≤ 𝑐 200𝑐 ≤ 30), and a fixed during the inference, i.e. assuming that if there is flattening,
flat prior of 0 ≤ 𝜂 ≤ 1. These priors reflect the large uncertainties in it is a constant density, flat core. We obtain good fits to the density
the potential halo masses hosting such dwarfs (e.g. Read et al. 2017; profiles with this assumption, but we remark that dwarf galaxies in
Jethwa et al. 2018; Nadler et al. 2020), and that in ΛCDM, such low- our regime are likely to have profiles shallower than NFW, but not
𝑀★ are not expected to form large cores (e.g. Di Cintio et al. 2014; actually flat (i.e. incomplete cusp-core transformation; Orkney et al.
Tollet et al. 2016; Lazar et al. 2020; Orkney et al. 2021). To remain 2021, Figure 1). We leave to future work an explorative inference with
consistent with Read et al. (2016b, 2017, 2019), we keep 𝜅 = 0.04

MNRAS 000, 1–18 (2024)

 18 M. P. Rey et al.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B7. Same as Figure 4 but showing ‘Halo 624 GM: Higher final mass’ at 𝑡 = 5.9 Gyr. Again, the H i distribution is visually flattened (right panels)
and with a clearly peaking 𝑣 𝜙,𝑔 profile (left). Although the H i in the centre is cold (top, middle) the interpretation of the rotation curve and 𝑣rot, H i remains
challenging (left).

the 𝜅 parameter left free, and an investigation of its degeneracies with

other parameters.
Figure D1 shows the result of the inference, with Fig 8 showing
the marginalized dark matter density profiles. The mass and concen-
tration are poorly constrained in both cases, although compatible at
1𝜎 with their true values (marked by vertical dashed lines). This is to
be expected as rotation curves extending to 𝑁H i ≥ 5 × 1019 cm −2 are
still in the rising phase, making it difficult to break degeneracies be-
tween higher-mass haloes with lower concentration and lower-mass
haloes with higher concentrations. We checked that this degeneracy
is strongly alleviated if providing the full rotation curve (𝑣 circ in
Fig 5) over an extended radius range (𝑁H i ≥ 5 × 1017 cm −2 ).
The star-forming galaxy has a marginal preference for a core (𝜂 =
0.57+0.27
−0.32
) that makes the inferred profile lower than its simulated
truth at 𝑟 ≤ 𝑟 1/2,3D (Figure 8). However, the posteriors remain wide
and parameters are poorly constrained compared to their input priors.
And, if observed, the visibly disturbed H i distribution of this object
(Figure 4) would likely make any inference on the absence or presence
of cores inconclusive.
The mass and concentration are similarly poorly constrained for
the quiescent dwarf (brown contours), but by contrast, the presence
of a steep cusp is confidently preferred (𝜂 = 0.08+0.35 −0.02
). As we
emphasize in Section 5, even with current instruments that can only
realistically capture the inner rise of the rotation curve in such faint
systems, this is enough to recover a steep cusp that would put strong
bounds on alternative models of dark matter.

This paper has been typeset from a TEX/LATEX file prepared by the author.

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 19

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure B8. Same as Figure B7 but at 𝑡 = 13.5 Gyr. The H i distribution is fairly regular and comparable in size to the stellar distribution, showcasing an
exponential H i profile. The H i is warm, with thermal support dominating 𝑣 𝜙,𝑔 , but the pressure-corrected 𝑣rot, H i is an accurate predictor of 𝑣circ in the very
inner part. A feedback event likely just occurred and started dispersing the H i disc, with its bubble visible in H i (top, right).

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20
Tg (K)

103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI
(M kpc 2 )

rD=
100

106
pc

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure C1. Same as Figure 5 but at 𝑡 = 11.9 Gyr. The similarities between the two figures showcase the long-term stability of the clear and easy to interpret rotation
curve in ‘Halo 624’. Feedback from old stellar populations (SNIa, AGB stars; see Rey et al. 2020) still inject energy into this small system (𝑣circ ≈ 10 km s −1 )
and can temporarily disrupt the gas flows (e.g. top, right; see also Figure 3).

MNRAS 000, 1–18 (2024)

 20 M. P. Rey et al.

25
Vcirc, tot cs
v ,g turb, g
104
Vrot, HI
20

Tg (K)
103
Velocity (km s 1 )

15 NHI 3 × 1019 cm 2
r1/2, 3D 1 kpc

107 Gas
10 HI

(M kpc 2 )

rD=
100
106

pc
5

1 kpc
5
0 0.2 0.4 0.6 0.8 1.0 10 0.2 0.4 0.6 0.8 1.0
Radius (kpc) Radius (kpc) 1018 1019 1020 1021
NHI (cm 2 )
Figure C2. Same as Figure 5 but at 𝑡 = 13.4 Gyr, showcasing that the rotation curve observed at 𝑧 = 0 has been in place for the last 500 Myr (≈ 5 full orbits at
300 pc).

MNRAS 000, 1–18 (2024)

 The smallest H i rotation curves 21
Halo 600
Halo 624

20
c200c

15

10

0.8

0.6
eta

0.4

0.2

108 109 1010 10 15 20 0.2 0.4 0.6 0.8

M200c c200c eta

Figure D1. Marginal distributions for the halo mass, halo concentration and
core size compared to 𝑟1/2,3D inferred from the star-forming and quiescent
dwarfs (blue and brown contours, respectively). Halo masses and concen-
trations are poorly constrained as our mock rotation curves only include the
inner rise. But this is enough to confidently infer a steep cusp in our quiescent
dwarf (𝜂 ≈ 0).

MNRAS 000, 1–18 (2024)