In this paper
AbstractIntroductionDataMUSEHSTAges and Chromosome MapKinematic Quality CutsMethodologyPopulation DefinitionsChromosome Map (Abundance Variations)Age–Metallicity Relation (AMR)Metallicity BinsKinematic ModelingProfiles and LikelihoodsJoint 2D Fits and Kinematic EllipticityResultsChromosome Map PopulationsMetallicity BinsAge–Metallicity PopulationsDiscussionThe P1/P2 dichotomyHelium abundances in the chromosome-map populationsAgreement with dynamical models and larger-radius studiesRelaxation and Dynamical EvolutionComparison with Typical Globular ClustersFormation ScenariosObservational ConstraintsScenario 1: Hierarchical AssemblyScenario 2: In-situ Gas ProcessingDebris, Mass Budget, and TimingOne enrichment event or two?LimitationsConclusionsAcknowledgmentsAppendixSample Size per SubpopulationChromosome Map: 1D Kinematic ProfilesMetallicity Bin KinematicsAge–Metallicity Relation KinematicsRotation PosteriorsReferencesWorksPhysics and Astronomy
oMEGACat. XI. Kinematic Characterization of Omega Centauri's Subpopulations: A Flattened Primordial Component
Abstract
Omega Centauri ( Cen; NGC 5139), the likely stripped nucleus of a dwarf galaxy accreted by the Milky Way, is one of the few galactic nuclei resolvable into individual stars. Because of its long relaxation time, the kinematics of its stars may still retain memory of how this nucleus formed. We combine HST proper motions and MUSE line-of-sight velocities from the oMEGACat survey to compare populations defined by abundance variations (the chromosome map), metallicity, and age within the half-light radius. The chromosome-map division separates two dynamically distinct components. The primordial stars have a flattened dispersion field, with kinematic ellipticity (), which we recover independently in the line-of-sight velocities and the proper motions. The enriched stars are consistent with in the line-of-sight velocities and are no more flattened than the full sample in the proper motions, and the two components show no strong difference in rotation. Metallicity bins show no comparable structure, because each bin mixes stars from both components, and the age-metallicity sample covers only the cluster core, where it cannot test for the flattening. The present-day central concentration of the most enriched (P2) stars suggests that they formed more centrally concentrated than the primordial stars, independent of assembly scenario. These results favor a picture in which the primordial stars retain the flattened orbital structure of the progenitor, while the enriched stars fell inward as dense clumps or formed in compact, centrally concentrated starbursts.
Introduction
Omega Centauri ( Cen; NGC 5139) is the most massive globular cluster in the Milky Way (MW), with a mass of approximately – (H. Baumgardt & M. Hilker 2018; N. Dickson et al. 2023). Unlike typical globular clusters (GCs), Cen exhibits complex stellar populations, spanning more than 1 dex in metallicity (C. I. Johnson & C. A. Pilachowski 2010; M. S. Nitschai et al. 2023, 2024), and an age spread of Gyr (S. Villanova et al. 2014; C. Clontz et al. 2024). These char-
Corresponding author: Peter J. Smith
Email: [email protected] acteristics strongly suggest that Cen is the remnant nuclear star cluster (NSC) of a dwarf galaxy that was accreted and stripped by the MW (Y.-W. Lee et al. 1999; K. Bekki & K. C. Freeman 2003; N. Neumayer et al. 2020; J. Pfeffer et al. 2021; S. O. Souza et al. 2026). Furthermore, recent astrometric detections of fast-moving central stars support the presence of an intermediate-mass black hole (IMBH) of at its core (M. Häberle et al. 2024a), a hallmark of galactic nuclei. Understanding the formation and evolution of Cen can therefore constrain the processes governing NSC formation and the assembly history of the MW (e.g., the Gaia-Enceladus/Sausage (GSE) merger; D. Massari et al. 2019; G. Limberg et al. 2022, which C. F. P. Laporte & M. D. A. Orkney 2026 date to Gyr ago and associate with a coeval burst of GC formation).
Decades of photometric and spectroscopic studies have revealed that Cen hosts multiple, distinct populations of stars (e.g., J. E. Norris & G. S. Da Costa 1995; Y.-W. Lee et al. 1999; L. R. Bedin et al. 2004; C. I. Johnson & C. A. Pilachowski 2010; A. F. Marino et al. 2011; A. Bellini et al. 2017). The oMEGACat survey1 (M. S. Nitschai et al. 2023; M. Häberle et al. 2024b) has since mapped the metallicity (M. S. Nitschai et al. 2024), age (C. Clontz et al. 2024), helium enhancement (C. Clontz et al. 2025), and chemical abundance (Z. Wang et al. 2026) distributions of these populations, as well as the intracluster medium (Z. Wang et al. 2025).
Using the pseudo-color “chromosome map” (ChM) (A. P. Milone et al. 2015, 2017) constructed from the combination of the F275W, F336W, F435W and F814W Hubble Space Telescope (HST) filters, the cluster separates into three primary groups: the primordial (P1) stars, and the chemically enriched intermediate (Im) and P2 stars (C. Clontz et al. 2026; E. Dondoglio et al. 2026), where we follow the naming of C. Clontz et al. (2026). The P2 population is characterized by extreme helium enhancement (), nitrogen and sodium enrichment, and oxygen depletion (J. E. Norris 2004; G. Piotto et al. 2005; C. I. Johnson & C. A. Pilachowski 2010; A. F. Marino et al. 2011, 2012; C. Clontz et al. 2025; Z. Wang et al. 2026). By linking the chromosome map on the red giant branch (RGB) to the subgiant branch (SGB) age–metallicity relation (AMR), C. Clontz et al. (2026) demonstrated a Gyr age gap between the P1 populations and the P2 populations across all metallicities, which requires an extended formation history (e.g., D. Romano et al. 2007, 2010; A. F. Marino et al. 2012; A. Bellini et al. 2017). The true population count
is likely even finer than this three-way split: A. Bellini et al. (2017) identify at least 15 subpopulations along the main sequence (MS), F. Akbaba et al. [2] recover ten distinct populations from clustering in abundance space, while C. Clontz et al. (2026) identify 14 subpopulations through data-driven clustering in photometry and metallicity.
The labels P1 and P2 echo the primordial and enriched populations of typical GCs, but differ from them in important ways: P2’s helium enhancement lies at the extreme of the range seen in GCs, and both P1 and P2 span a wide range in [Fe/H] (C. I. Johnson & C. A. Pilachowski 2010; A. F. Marino et al. 2011) and in age (C. Clontz et al. 2026), whereas the populations of typical clusters share a single metallicity and are close to coeval (N. Bastian & C. Lardo 2018). The scenarios proposed for the multiple populations of typical clusters therefore may not apply to Cen’s P1 and P2.
Stellar kinematics probe the assembly history of stellar systems. In particular, the multiple stellar populations observed in typical GCs have long been known to exhibit distinct spatial distributions (e.g., C. Lardo et al. 2011; E. Leit inger et al. 2023; V. J. Mehta et al. 2025) and kinematic properties (e.g., M. J. Cordero et al. 2017; A. P. Milone et al. 2018; M. Libralato et al. 2023; E. Dalessandro et al. 2024; G. Cordoni et al. 2025; E. I. Leitinger et al. 2025). These differences can provide clues about the formation mechanisms of the various populations, such as whether they formed in situ or were accreted from external sources (e.g., pristine gas from the surroundings, A. D’Ercole et al. 2008; or infalling star clusters, [4]).
Standard models predict that P2 forms from enriched ejecta left behind by P1 stars, which settle into a compact, rotating, centrally concentrated configuration before undergoing star formation (K. Bekki 2010; A. Mastrobuono-Battisti & H. B. Perets 2013). Hydrodynamical simulations of second-generation formation from asymptotic giant branch (AGB) ejecta mixed with accreted low-density pristine gas likewise produce a compact, centrally concentrated second generation (A. Yaghoobi et al. 2022). The specific enrichment channel is debated, with proposed sources including AGB stars (e.g., A. D’Ercole et al. 2008; F. D’Antona & P. Ventura 2026), interacting binaries (S. E. de Mink et al. 2009), fast-rotating massive stars (T. Decressin et al. 2007), and supermassive or accreting extremely massive stars (P. A. Denissenkov & F. D. A. Hartwick 2014; M. Gieles et al. 2018, 2025) (see A. Renzini et al. 2015; N. Bastian & C. Lardo 2018, for critical assessments), and most predict that P2 is more centrally concentrated than P1, since enriched ejecta settle toward the cluster center. Recent observations, however, present a more mixed view: among dynamically young clusters, some have a centrally concentrated enriched population, some a centrally concentrated primordial population, and some fully mixed populations (E. Leitinger et al. 2023).
Large proper-motion and MUSE samples now show that kinematic differences between the populations are common. In most clusters the P2 stars have enhanced radial anisotropy compared to the P1 stars, which is consistent with the P2 stars being more centrally concentrated (A. P. Milone et al. 2018; M. Libralato et al. 2023; E. Dalessandro et al. 2024; G. Cordoni et al. 2025). The evidence for a rotation difference is similarly mixed, with no universal, consistent difference in rotation between P1 and P2 stars across clusters (M. J. Cordero et al. 2017; E. Dalessandro et al. 2024; E. I. Leitinger et al. 2025). These differences are weaker in dynamically older clusters (E. Dalessandro et al. 2024; G. Cordoni et al. 2025), and some clusters show no significant difference or the opposite sense (e.g., S. Martens et al. 2023).
The standard formation models were developed in the context of GC formation, not NSCs embedded within a dwarf galaxy potential, so their applicability to Cen is not guaranteed. Nuclear star clusters commonly rotate in the same sense as their host galaxies, and some are significantly more flattened and rapidly rotating than GCs (N. Neumayer et al. 2020). The host galaxy likely imposes large-scale angular momentum, meaning P1 itself may have formed in a flattened configuration, unlike the isotropic primordial stars assumed in standard multiple population formation models for GCs. Crucially, different formation pathways are expected to leave distinct kinematic imprints, including differences in population-dependent spatial flattening, that can survive the long-term dynamical evolution of old clusters and thereby allow different scenarios to be distinguished observationally (A. Mastrobuono-Battisti & H. B. Perets 2013; V. Hénault-Brunet et al. 2015; A. Mastrobuono-Battisti & H. B. Perets 2016; M. J. Cordero et al. 2017) (see also M. A. Tiongco et al. 2019; E. Vesperini et al. 2021; F. I. Aros et al. 2025). In Cen, each episode of star formation or accretion may likewise have left a residual kinematic signature.
Cen is particularly well suited to this search, because of its long relaxation time (the half-mass relaxation time is 21 Gyr; H. Baumgardt & M. Hilker 2018; H. Baumgardt & E. Vasiliev 2021, updated online catalog),2 which allows it to retain memory of its initial conditions for longer periods compared to typical globular clusters (A. Mastrobuono-Battisti & H. B. Perets 2013). This makes it possible to observe kinematic differences between populations that may have formed under different conditions or at different times.
Previous studies of Cen have revealed complex kinematic structures, including a kinematically cold, disk-like component at – (G. van de Ven et al. 2006), and a counter-rotating core (R. Pechetti et al. 2024). Beyond the half-light radius (), population-resolved studies find that the primordial stars rotate faster than the enriched stars at (A. Bellini et al. 2018), and have a broader angular-momentum distribution and weaker radial anisotropy at – (T. Ziliotto et al. 2026). Out to large radii, aluminum (Al)-poor stars are also closer to isotropy than Al-rich stars, although the two groups rotate alike and do not map one-to-one onto P1 and P2 (G. Pagnini et al. 2026). Recently, M. Häberle et al. (2025) provided a comprehensive 3D kinematic analysis of the inner half-light radius of the cluster using the oMEGACat dataset, investigating the rotation profile, velocity anisotropy, and degree of energy equipartition in the cluster. M. Häberle et al. (2026) extended the line-of-sight (LOS) rotation curve to with integrated-light spectroscopy from the SDSS-V Local Volume Mapper (N. Drory et al. 2024; J. A. Kollmeier et al. 2026). While M. Häberle et al. (2025) looked for metallicity-dependent kinematics, they found no significant differences when splitting the sample into [Fe/H] bins, motivating the more detailed population definitions we use here.
In this paper, we compare the kinematics of the multiple populations of Cen within the half-light radius, using high-precision HST proper motions (PMs; M. Häberle et al. 2024b) and MUSE LOS velocities (M. S. Nitschai et al. 2023) from the oMEGACat survey. We divide the stars by chromosome-map position, by [Fe/H], and by position on the AMR, and model the rotation, dispersion, and anisotropy of each population in 1D and 2D. Section 2 describes the data and quality cuts, Section 3 the population definitions and kinematic models, Section 4 the results, Section 5 their implications for the formation of Cen, and Section 6 our conclusions.
Data
The basis of this work is the oMEGACat catalog, a joint HST+MUSE survey of Cen. The catalog combines HST photometry and proper motions from M. Häberle et al. (2024b) with MUSE spectroscopy from M. S. Nitschai et al. (2023), and each dataset covers the central half-light radius of the cluster ( \approx 7.6 pc; H. Baumgardt & M. Hilker 2018; H. Baumgardt & E. Vasiliev 2021, updated online catalog).
MUSE
The MUSE dataset consists of integral field spectroscopy covering the half-light radius of Cen. The MUSE data are processed with the PampelMuse software ((S. Kamann et al. 2013)), which extracts individual stellar spectra from the crowded field. Line-of-sight velocities and metallicities are measured from these spectra using the spexxy software ((T.-O. Husser et al. 2016)). The final catalog contains spectra for over 300,000 stars. Typical LOS velocity uncertainties for bright RGB stars are ((M. S. Nitschai et al. 2023)).
HST
The HST dataset provides high-precision proper motions and seven-band photometry (ranging from ultraviolet (UV) to near-infrared) within the half-light radius of Cen. The proper motions of 1.4 million stars are derived from multi-epoch HST imaging with a typical baseline of 20.1 yr. The proper motions are relative to the local mean motion of the cluster stars, which removes any plane-of-sky rotation, and achieve a median 1D precision of for mag, and down to in the highly sampled cluster core ((M. Häberle et al. 2024b)).
Ages and Chromosome Map
We additionally use stellar ages from C. Clontz et al. (2024), who derived precise ages for SGB stars in Cen by fitting the HST photometry with isochrones that adopt a specific C+N+O–[Fe/H] relation. The ages span a range mostly from 10 to 13 Gyr, with median uncertainties of 0.68 Gyr. We also use the chromosome map from M. S. Nitschai et al. (2024), described in § 3.1.1.
Kinematic Quality Cuts
To ensure reliable kinematic measurements, we apply strict quality cuts to the datasets, adopting the recommendations from M. Häberle et al. (2025) for both the HST proper motions and MUSE LOS velocities. We select stars with high-quality measurements, removing those with large uncertainties or potential contamination from neighboring sources. We apply their catalog flags selection_hq_astrometry_and_membership (HST proper motions and cluster membership) and selection_hq_los (MUSE LOS velocities). Every star therefore appears in both catalogs, so the single membership selection applies to both datasets. Table A1 (Appendix § 9) gives the number of stars in each subpopulation after each cut, and shows that no individual subpopulation is preferentially removed by the quality cuts.
| Chromosome map | ||||||
| Cut | P2 | Im | P1 | Total | ||
| Population defined | 2323 | 1187 | 1531 | 5041 | ||
| Good astrometry & membership | 2114 | 1079 | 1363 | 4556 | ||
| Good LOS velocity | 1904 | 940 | 1224 | 4068 | ||
| Retained | 82% | 79% | 80% | 81% | ||
| AMR | ||||||
| Cut | Tight | Diffuse | Metal-Rich | Total | ||
| Population defined | 3448 | 3313 | 1008 | 7769 | ||
| Good astrometry & membership | 3094 | 3024 | 847 | 6965 | ||
| Good LOS velocity | 2573 | 2496 | 653 | 5722 | ||
| Retained | 75% | 75% | 65% | 74% | ||
| Used in fits | 2573 | 2496 | – | 5069 | ||
| quintiles | ||||||
| Cut | Met 1 | Met 2 | Met 3 | Met 4 | Met 5 | Total |
| Population defined | 1899 | 1730 | 1727 | 1709 | 1723 | 8788 |
| Good astrometry & membership | 1575 | 1494 | 1464 | 1454 | 1411 | 7398 |
| Good LOS velocity | 1239 | 1237 | 1237 | 1238 | 1238 | 6189 |
| Retained | 65% | 72% | 72% | 72% | 72% | 70% |
Table A1. Number of stars in each subpopulation after each quality cut. “Retained” gives the fraction of the first row that survives both cuts. The Metal-Rich AMR group is excluded from the fits (§ 4.3).
At this point, the quality cuts are purely kinematic; we apply additional cuts in the following sections to ensure that the stars have the necessary chemical abundance and age information to be classified into the various subpopulations, each within the magnitude range in which the population is defined (§ 3.1).
Methodology
To characterize the kinematic properties of the multiple stellar populations in Cen, we first define subpopulations using three complementary methods based on photometric abundance variations, [Fe/H], and age. We then model the kinematics of each subpopulation using 2D spatial maps to infer their 3D kinematic structure. In the main text we focus on the 2D fields; the corresponding 1D radial profiles are given in Appendices B–D.
Population Definitions
We employ three complementary population divisions to test which aspect of Cen’s chemistry traces its dynamics: the chromosome map (star-by-star abundance variations), groupings on the AMR, and pure [Fe/H] quintiles. By comparing the kinematics of these populations, we can constrain the formation history of the cluster and the processes that shaped its multiple stellar populations.
Chromosome Map (Abundance Variations)
The chromosome map, introduced by A. P. Milone et al. (2015), separates stellar populations based on multi-band photometry sensitive to light-element abundance variations. It is built from two pseudo-colors, from and from , each measured relative to fiducial lines along the RGB ((G. Piotto et al. 2015; A. P. Milone et al. 2015; M. S. Nitschai et al. 2024)). In particular, the index is sensitive to carbon, nitrogen, and oxygen abundances through the molecular bands (OH, NH, CN, and CH features), while is primarily temperature-sensitive and traces helium enrichment indirectly through its effect on stellar temperature, while also tracing iron abundance, or total metallicity.
We adopt the chromosome map from M. S. Nitschai et al. (2024), with the population boundaries from C. Clontz et al. (2025), to define three distinct populations within Cen: primordial (P1) stars, which occupy the lower portion of the map and are characterized by primordial abundance patterns; enriched (P2) stars, which occupy the upper portion and exhibit enhanced nitrogen and helium; and an intermediate (I1) population that lies between P1 and P2. In Figure 1, we illustrate the definition of these populations on the chromosome map, showing the distinct regions occupied by P1, P2, and Im stars as well as the range of metallicities covered by each group. In addition to C, N, O, and He, the sequences also differ in Al and in heavy elements such as La (C. I. Johnson & C. A. Pilachowski 2010; A. F. Marino et al. 2011; A. C. Mason et al. 2026; Z. Wang et al. 2026). Full abundance and age characterizations of these groups are given in C. Clontz et al. (2025); Z. Wang et al. (2026); C. Clontz et al. (2026). This separation allows us to investigate the kinematic properties of these chemically distinct populations, without requiring detailed spectroscopic abundances for every star.
Figure 1. Chromosome map (ChM) of Cen RGB stars, constructed from the pseudo-color indices and from HST photometry (M. Häberle et al. 2024b). is the F275W−F814W color and the pseudo-color , each verticalized relative to fiducial lines bounding the RGB. (a) Stars are color-coded by MUSE metallicity [Fe/H] (M. S. Nitschai et al. 2023). (b) The three populations used in this work, P1 (red), Im (yellow), and P2 (blue), are defined: primordial (P1, lower sequence), intermediate (Im, middle), and enriched (P2, upper sequence), following M. S. Nitschai et al. (2024) and C. Clontz et al. (2025); dashed lines mark the P1/Im and Im/P2 boundaries. The vertical separation primarily traces nitrogen enhancement, while the horizontal axis is sensitive to temperature, meaning it probes largely metallicity (and helium enhancement) (A. P. Milone et al. 2015; C. Clontz et al. 2025).
Beyond the shared kinematic quality cuts (§ 2.4), stars are required to have a valid chromosome-map pseudo-color, and to additionally fall on the RGB within , following the RGB selection of C. Clontz et al. (2026); the faint limit keeps the sample away from the SGB. The final sample has stars: , , and .
Age–Metallicity Relation (AMR)
We use the AMR measured by C. Clontz et al. (2024) to define subpopulations based on stellar ages and iron abundances. The AMR traces the chemical evolution of Cen in time, allowing us to separate stars into groups that reflect different formation epochs and enrichment histories. We show in Figure 2 the distribution of stars in the AMR space. In panel (a), we show a 2D histogram, colored by the log-density of points in the diagram, showing
Figure 2. AMR for SGB stars in Cen, derived by C. Clontz et al. (2024) from isochrone fitting to HST photometry. (a) AMR colored by the number of stars per bin (log scale), showing the distinct sequences present. (b) The three AMR populations (Tight and Diffuse streams from C. Clontz et al. (2024), Metal-Rich group from C. Clontz et al. (2026); see § 3.1 for their use in this work) are indicated: “Tight” stars (the younger, narrower sequence at fixed metallicity), “Diffuse” stars (older at fixed metallicity, broader scatter), and “Metal-Rich” stars (the most iron-abundant portion, generally younger). Dashed lines mark the population boundaries: stars above [Fe/H] = (gray shading) form the Metal-Rich group, which we exclude from the fits, and the diagonal line separates the Tight and Diffuse sequences. Median age uncertainties are Gyr.
ing the distinct sequences present. In panel (b) we define three populations based on their position in the AMR space: “Tight” stars (in green), which, at fixed metallicity, are the youngest and occupy a narrow linear sequence; “Diffuse” stars (in orange), which are older at fixed metallicity and exhibit a broader spread; and “Metal-Rich” stars (in gray), which occupy the most metal-rich portion of the AMR and are generally younger. This classification allows us to investigate how age and iron abundance correlate with kinematic properties. AMR membership additionally requires a non-null fiducial age and [Fe/H] from C. Clontz et al. (2024), restricting this sample to SGB stars with well-measured ages. After the quality cuts, the samples contain 2573 Tight, 2496 Diffuse, and 653 Metal-Rich stars, i.e., a total of stars.
Metallicity Bins
The simplest way to define subpopulations is to divide the sample into bins based solely on bulk metallicity [Fe/H].3 This approach lets us test whether metallicity alone drives kinematic differences among the stars in Cen. We divide the RGB sample into quintiles of [Fe/H], creating five distinct metallicity bins that span the full metallicity range observed in the cluster. In Figure 3, we illustrate the definition of these metallicity bins using a kernel density estimate (KDE) of the [Fe/H] distribution, indicating the boundaries of each quintile with black vertical lines. These stars are required to have (the RGB sample of M. S. Nitschai et al. (2024) has a mean metallicity uncertainty of dex) and a non-null [Fe/H] measurement, giving ; the high-quality-astrometry cut reduces this to , and the high-quality LOS cut applied in all analyses (§ 2.4) further reduces it to , split into five quintile bins of stars each.
Figure 3. Kernel density estimate (KDE) of the MUSE metallicity [Fe/H] distribution for all RGB stars used in this work to construct the metallicity bins. The five colored regions indicate the quintiles of the distribution, which are used to define the metallicity bins for kinematic analysis. The bins are labeled Met 1 (most metal-poor) to Met 5 (most metal-rich). The metallicity distribution spans dex and peaks around [Fe/H] , consistent with the full metallicity distribution of Cen as reported by M. S. Nitschai et al. (2024).
We now model the 2D kinematics of each of these populations (§ 3.2).
Kinematic Modeling
We characterize the kinematic structure of each subpopulation using 2D spatial maps of the velocity dispersion, rotation, and velocity anisotropy, from which we infer aspects of the underlying 3D geometry and identify kinematic differences that may reflect their formation histories. The corresponding 1D radial profiles, fit with both discrete (unbinned) and binned methods, are presented in Appendix B as a check for trends and correlations with radius.
Profiles and Likelihoods
We model the velocity dispersion using a Plummer dispersion profile (Equation (1)), which describes a centrally peaked dispersion that flattens at large radii and matches our binned profiles (Appendix § 10). The rotation is modeled as a Lynden-Bell rotation profile (e.g., D. Lynden-Bell 1967; A. L. Varri & G. Bertin 2012; V. Hénault-Brunet et al. 2015; S. Kamann et al. 2018, a rotation curve that rises linearly in the core and declines outward, as expected both after violent relaxation (D. Lynden-Bell 1967) and for quasi-relaxed rotating clusters (A. L. Varri & G. Bertin 2012). We implement all our models in JAX (James Bradbury et al. 2018) and sample the posterior distributions of the model parameters using the No-U-Turn Sampler (NUTS) implemented in the numpyro package (D. Phan et al. 2019).
The Plummer dispersion profile is given by:
where is the central velocity dispersion (km s for LOS fits, mas yr for PM fits), is the scale radius (arcsec), and is the projected radius (arcsec) from the cluster center.
The Lynden-Bell rotation profile is given by:
where is the projected peak rotation amplitude (km s or mas yr), is the radius (arcsec) at which the rotation curve reaches its maximum, and is the projected distance (arcsec) along the kinematic major axis. The model rises linearly at small and falls as at large . When fitting our models to the data, we adopt Gaussian likelihoods, computing the mean velocity and velocity dispersion at the location of an individual star. The total likelihood is then the product of the likelihoods of all the individual stars, though in practice, we compute the log-likelihood and sum over the stars:
where is the observed velocity of star , is the model mean velocity at the location of star , is the model velocity dispersion at that location, and is the measurement uncertainty for star .
Joint 2D Fits and Kinematic Ellipticity
For each population and velocity component (LOS, , ), we also construct 2D spatial maps of the velocity dispersion, the LOS rotation, and anisotropy ratio . This allows us to fit for any elliptical distortion in the kinematic fields, which may be blurred or averaged out in the 1D profiles, and constrain the azimuthal dependence of the velocity anisotropy. Because the likelihood function evaluates the kinematics at the location of each star, the spatial selection function no longer plays a role, allowing us to infer unbiased kinematic fields despite the uneven data coverage from the survey footprint. The ellipticity parameter is therefore derived purely from the kinematic amplitudes that vary as a function of azimuth, not from the spatial distribution of the observed stars.4
We fit the LOS, , and velocities of each population jointly, with a single kinematic axis whose position angle (PA) is free and is set mainly by the LOS rotation.5 Each component has its own Plummer profile and ellipticity along this axis. We allow to take either sign (flat prior on ), with describing contours elongated perpendicular to the axis. We adopt this somewhat unusual parametrization because a prior bounded at will be biased toward positive even when the data are drawn from a perfectly spherical distribution. This allows for posteriors that are consistent with zero flattening to cluster around , allowing for easier interpretation of the results.
We measure the evidence for flattening with the Savage–Dickey Bayes factor (J. M. Dickey 1971; R. Trotta 2008) of the elliptical model against . Because a shallow dispersion profile turns a modest azimuthal contrast into a large , we also quote at , which provides a more intuitive measure of the flattening of the velocity field.
The 2D extension of the Plummer dispersion profile allows for an elliptical distortion by replacing the circular radius with an effective radius that accounts for the ellipticity and PA of the kinematic field. The 2D dispersion model is then given by:
where is the effective radius that accounts for the ellipticity and PA of the kinematic field, where and are the coordinates aligned with the major and minor axes of the ellipse (rotated from sky coordinates by the kinematic position angle PA, as defined in Equations (5)–(6) below), and is the kinematic ellipticity.
Likewise, for the rotation field, we allow for an elliptical distortion by replacing the circular radius with the effective radius in the rotation model:
with the same definition of as above, but with its own ellipticity, ; is the projected (LOS) rotation velocity of the cluster. The rotation curve of Cen peaks near the edge of our field, so and are degenerate in our fits. To compare the rotation of the populations, we therefore evaluate the rotation velocity on the kinematic major axis at , , for each posterior draw.
In the joint fits the LOS rotation field fixes the direction of the shared axis, which removes the PAPA+ ambiguity of a signed ellipticity, so we adopt a flat prior on the PA. We adopt large, uninformative priors on all other model parameters. In the joint fits we fit a constant mean for each PM component, which absorbs any residual offset in the relative proper motions, and fit the LOS systemic velocity as a free constant in the same way. We fit no PM rotation term because the proper motions measured by HST are relative to the local cluster motion and therefore do not capture the motion of the cluster or streaming motion in the tangential direction, only the dispersion about the local mean motion.
To validate that the kinematic signatures recovered by our model are not artifacts of the parametric fitting approach, we compare our anisotropy maps to the non-parametric Voronoi-binned maps of M. Häberle et al. (2025), using the same sets of high-quality stars. Our models produce maps that are in good agreement with the Voronoi-binned maps, indicating that the kinematic features we identify are not dependent on the specific modeling approach.6
Results
We present the kinematic results for the various subpopulation definitions outlined in § 3.1. We begin with the chromosome-map populations, which reveal the most distinct kinematic differences. We then show that populations defined by pure metallicity bins recover no comparable structure, and describe why the AMR sample, which has limited radial coverage, cannot test for the P1 flattening.
Chromosome Map Populations
The chromosome-map populations show the clearest differences in kinematic structure (Figure 4 and Figure 5). Figure 6 shows the posterior distributions of the elliptical fits directly: in the LOS, the P1 ellipticity posterior is offset from zero, from the bulk, and from the P2/Im posteriors. In , P1 is again the most flattened population and is offset above the bulk, but its posterior overlaps those of Im and P2, which are consistent with the bulk sample. The PM contrast is smaller because the whole cluster is flattened in () but not in : the cluster rotation, seen in projection, flattens the field of every population ((M. Häberle et al. 2025)), so the additional flattening of P1 is measured against an already flattened baseline, whereas in the LOS the bulk, Im, and P2 are close to round. In 1D, the Im and P2 populations display largely similar dispersion and rotation profiles (Appendix § 10), while the P1 stars exhibit a shallower LOS dispersion profile (Figure B1); the azimuthally averaged PM anisotropy profiles of all three populations agree within the uncertainties (Figure B2).
Figure 4. HST proper-motion kinematic maps for the bulk chromosome-map sample (All), P2, Im, and P1 (columns, left to right). Rows, top to bottom: radial PM dispersion , tangential PM dispersion , and anisotropy ratio . The dashed regions indicate the convex hull containing the data in each subset. The inset at the top left of each panel marks the cluster major (solid) and minor (dotted) axes. Thin black lines in the dispersion panels are iso-dispersion contours, evenly spaced across the color range. The anisotropy maps are the ratio of the fitted and fields in the rows above. For P1 (panel l), the flattened and round fields combine into a quadrupole pattern: near-isotropic along the minor axis and radially anisotropic () along the major axis (see § 4.1). Im and P2 anisotropy maps show no extended quadrupole and are near-isotropic (P2) or mildly tangential (Im) in the core, becoming radially anisotropic at larger radii in all directions.
Figure 5. 2D MUSE LOS kinematic maps for the bulk chromosome-map sample (All), P2, Im, and P1 (columns, left to right). Rows, top to bottom: LOS rotation velocity and LOS velocity dispersion . Dashed lines indicate the convex hull containing the data in each subset. The inset at the top left of each panel marks the cluster major and minor axes, as in Figure 4. Thin black lines in the dispersion panels are iso-dispersion contours, evenly spaced across the color range. The P1 dispersion field (panel h) is elliptical (; see § 4.1). The Im and P2 dispersion fields are consistent with azimuthally symmetric distributions.
Figure 6. Posterior distributions of the dispersion-field ellipticity for the P2, Im, and P1 chromosome-map populations, from the elliptical fits to (a) , (b) , and (c) , compared against the bulk chromosome-map sample posterior (median as dashed line, 68% interval as gray band). Horizontal bars mark each posterior’s median and full range. Negative describes a dispersion field elongated perpendicular to the kinematic axis (§ 3.2). P1 is flattened in and (Bayes factors and 30), while Im and P2 are consistent with in the LOS. In , the bulk sample is also flattened () and the Im and P2 posteriors include . In , all posteriors include . The LOS rotation posteriors are shown in Figure E1.
Figure B1. 1D LOS kinematic profiles for the chromosome-map populations P1 (red), Im (yellow), and P2 (blue) and the bulk sample (black), as a function of projected radius. (a) LOS velocity dispersion , with best-fit Plummer model and posterior bands. (b) LOS rotation velocity from the Lynden-Bell model. (c) Position angle of the rotation. In all panels, we additionally show the binned version of the profiles for comparison. The P1 dispersion profile is lower in the core and flatter with radius than those of Im and P2, while the rotation axis of Im may be offset by from that of P2 (§ 4.1), and P1 rotates more slowly.
Figure B2. 1D proper-motion kinematic profiles for the chromosome-map populations P1 (red), Im (yellow), and P2 (blue) as well as the bulk sample (black), as a function of projected radius. (a) Radial PM dispersion . (b) Tangential PM dispersion . (c) Anisotropy ratio . All populations agree within the uncertainties.
Within , P1 rotates no faster than Im and P2, and somewhat more slowly in projection (Figure 5 and Figure B1). At , the elliptical fits give for P1, against (P2) and (Im) (Figure E1d). The peak amplitudes overlap (Figure E1a), because the peak of the rotation curve lies at the edge of our field and trades off against (§ 3.2). The axis of rotation, however, may differ for Im. The elliptical fits place its kinematic position angle (the line of maximum rotation) at , against for P1 and for P2 (Figure E1j). Fitting the individual LOS velocities of each population with only a constant and a dipole in position angle, , which assumes no rotation curve, dispersion profile, or flattening, places the Im axis from that of P2. Because the rotation signal grows with radius, both measurements are set mostly by the outer stars.7
Figure E1. Posterior distributions of the LOS rotation parameters for the chromosome-map populations (left), the [Fe/H] quintiles (middle), and the Tight and Diffuse AMR populations (right). Rows, top to bottom: projected peak rotation amplitude (a–c), rotation velocity on the kinematic major axis at derived from the same posterior draws (d–f), rotation-field ellipticity (g–i), and position angle of the kinematic major axis, east of north (j–l). Bulk posteriors and bars as in Figure 6.
In 2D, the kinematic maps reveal clear differences in the shape of the dispersion fields. When we consider the azimuthal dependence and flattening of the dispersion fields, the P1 stars exhibit kinematic flattening along the major axis in and ( and , with Bayes factors and 30; Figure 6, with maps in Figure 4 and Figure 5), compared to the spherical dispersion fields of the Im and P2 stars (Figure 5 and Figure 6).
For P1, the elongated field and the round field combine into a quadrupole pattern: along the minor axis and along the major axis. The Im and P2 stars show no extended quadrupole: their maps are near-isotropic (P2) or mildly tangential (Im, ) in the core and become radially anisotropic beyond in all directions. M. Häberle et al. (2025) identified quadrupole-like “plumes” of the same form in the PM anisotropy map of the whole cluster (their Figure 11), which they attribute to the physical anisotropy combined with the projection of the cluster rotation.
The P1 ellipticities correspond to in both components at , consistent with the contrast we measure without a parametric model.8 At the same radius, Im and P2 give and in the LOS and and in , and the bulk sample and . Im and P2 show no significant flattening in any component (; Figure 6). Kinematic ellipticity is independently detected in both and , as these two components are measured from independent datasets,
HST astrometry and MUSE spectroscopy, though in the same sample of stars (Figure 6).
We interpret the apparent difference between the slightly shallower 1D dispersion profile (Figure B1) and the elliptical 2D field (Figure 5) as a result of the data footprint, which is elongated and nearly aligned with the cluster geometry: the majority of stars at larger radii fall along the flattened direction, which would bias the 1D profiles toward higher dispersion at large radii and thus a shallower radial gradient than the 2D field.
In summary, the chromosome map isolates a single dynamically distinct component, the primordial stars, which show a flattened dispersion field in both datasets (Figure 6), with the PM anisotropy quadrupole that the flattening implies (Figure 4), and rotation no faster than that of Im and P2 within (Figure 5). Having established that the chromosome map does reveal dynamically distinct kinematic components, we next test whether simpler metallicity-only divisions recover any comparable structure.
Metallicity Bins
We show the definition of the five [Fe/H] quintiles in Figure 3 and their ellipticity posteriors in Figure 7. The 1D LOS and PM profiles are presented in Figure C1 and Figure C2 (Appendix C). The metallicity quintiles show consistent LOS and PM dispersion profiles across all bins (Figure C1 and Figure C2), no discernible trend in rotation with metallicity (Figure C1), and PM anisotropy curves that agree within the uncertainties (Figure C2). None of the quintile ellipticity posteriors is a secure detection, and they show no coherent trend with [Fe/H]. The only secure flattening we recover is when all populations are combined. Figure 7 shows this directly: the ellipticity posteriors of the individual [Fe/H] quintiles overlap both each other and the bulk-cluster posterior in both the LOS and PM fits.
!Three violin plots showing dispersion ellipticity for [Fe/H] quintiles in LOS, PM radial, and PM tangential dispersion bbox=0.08,0.08,0.92,0.28
Figure 7. As in Figure 6, but for the [Fe/H] quintiles, compared against the bulk [Fe/H] sample posterior. Unlike the chromosome-map decomposition, no [Fe/H] quintile is preferentially offset from the bulk posterior in any component.
Figure C1. 1D LOS kinematic profiles for the five [Fe/H] quintile populations (colored from most metal-poor to most metal-rich) and the bulk sample (black). (a) LOS velocity dispersion . (b) LOS rotation velocity . (c) Position angle of the rotation. No significant differences in dispersion or rotation are detected across the quintiles.
Figure C2. 1D proper-motion kinematic profiles for the five [Fe/H] quintile populations (same color scheme as Figure C1). (a) Radial PM dispersion . (b) Tangential PM dispersion . (c) Anisotropy ratio . Dispersion and anisotropy profiles are consistent across all metallicity quintiles within the uncertainties.
We interpret this as following directly from the substantial overlap in the [Fe/H] distributions of P1, Im, and P2 ((C. Clontz et al. 2026; Z. Wang et al. 2026)): each [Fe/H] bin contains a mixture of the kinematically distinct chromosome-map populations, and therefore dilutes their signatures.
In summary, we find no coherent departure from the bulk cluster in ellipticity, rotation, or anisotropy for any of the five [Fe/H] quintiles, and no trend with [Fe/H] (Figure 7). We next test the more structured AMR divisions.
Age–Metallicity Populations
The AMR sample ( and ; we exclude the small, , Metal-Rich sequence) is limited to the central region with reliable F606W coverage: ninety percent of its stars lie within 139″ of the center, inside the core radius ( pc ; H. Baumgardt & M. Hilker 2018, updated online catalog). Inside the core the dispersion field is nearly flat, so it constrains the ellipticity only weakly, and every AMR posterior, including that of the bulk AMR sample, is broad (Figure D5). With this coverage, the AMR populations do not allow a meaningful comparison of kinematic structure with the chromosome-map or metallicity divisions. We show their ellipticity posteriors and 1D profiles in Appendix D.
Figure D5. As in Figure 6, but for the Tight and Diffuse AMR populations, compared against the bulk AMR sample posterior. Because the AMR sample lies almost entirely inside the core radius, where the dispersion field is nearly flat, every posterior, including that of the bulk, is broad and weakly constrained (§ 4.3).
Discussion
The P1/P2 dichotomy
Within ω Cen’s half-light radius, the chromosome-map primordial (P1) and enriched (Im, P2) populations are dynamically distinct (§ 4.1). P1’s dispersion field is flattened along the cluster major axis, with kinematic ellipticity recovered independently from HST proper motions and MUSE LOS velocities (Figure 6), and because its field is round, its proper-motion anisotropy follows a quadrupole pattern (Figure 4), near-isotropic along the minor axis and radially anisotropic along the major axis. Im and P2 are consistent with spherical symmetry in the LOS dispersion field and are no more flattened than the bulk sample in the proper motions (§ 4.1), show no extended quadrupole, and transition from near-isotropy or mild tangential anisotropy in the core to radial anisotropy at intermediate radii. Because P1 rotates no faster than P2/Im within , and more slowly at 200″, while its LOS flattening and extended quadrupole appear in P1 alone, these features are not a projection of the global cluster rotation. Neither the flattening nor the quadrupole appears when the cluster is divided by metallicity (§ 4.2), so the star-by-star abundance variations traced by the chromosome map, not bulk metallicity, separate the dynamically distinct components of ω Cen. [Fe/H] and AMR position may trace gas supplied by the host galaxy ((C. F. P. Laporte & M. D. A. Orkney 2026)), whereas the chromosome map traces enrichment within the cluster. Divisions by [Fe/H] or by position on the AMR mix the chromosome-map populations: among stars with both AMR and chromosome-map labels from C. Clontz et al. (2026), the Tight AMR sequence is 21% P1, 34% Im, and 45% P2, and the Diffuse sequence is 47%, 20%, and 33%. The Diffuse sequence is more P1-rich than the bulk chromosome-map sample, but the AMR sample covers only the core and cannot test whether this mixture carries the P1 flattening (§ 4.3).
We find that P2 is also more centrally concentrated than P1: in C. Clontz et al. (2026)’s stacked, data-driven clustering sample (Figure 8), the P2 fraction declines with projected radius while the P1 fraction rises, as seen even more clearly in studies covering a larger radial range (e.g., A. Bellini et al. 2009; M. Scalco et al. 2024; E. Dondoglio et al. 2026). Because relaxation would only act to erase this difference, the observed radial trend indicates that P2 formed more centrally concentrated than P1, and that the two populations have not yet fully mixed. P2 was probably also denser than P1 when it formed, as most formation scenarios assume (e.g., F. I. Aros et al. 2025), and is now relaxing toward a more extended configuration.
Figure 8. Fraction of P1 (red), Im (yellow), and P2 (blue) stars as a function of projected radius, using C. Clontz et al. (2026)’s stacked, data-driven clustering sample. P2 is the dominant population in the cluster core and its fraction declines outward, while the P1 fraction rises with radius, as does that of Im. (a) Fractions of each population in radial bins; dashed lines are linear fits and shaded bands their intervals. (b) The enriched fraction (P2+Im) from this work (two-color symbols) compared with the enriched (2P) fraction of E. Dondoglio et al. (2026) (gray; their Figure 5), which they compute relative to their canonical 1P+2P stars only, excluding their iron-rich AI and AII groups, and whose outermost bin is centered at and reaches ; the black dashed line and gray band show a joint linear fit to both datasets. Slopes in the legends are in fraction per arcmin; the P2, P1, and Im slopes differ from zero at , , and .
The main difference we detect lies in the distribution of orbits. In the self-consistent rotating models of A. L. Varri & G. Bertin (2012), the density and dispersion fields are flattened mainly by rotation, and a faster-rotating model is a flatter one. P1 is the flatter component despite rotating more slowly, which suggests that the two components differ in orbital structure. Within we measure the shape of this structure, a flattened dispersion field. The flattening is consistent with P1 orbits whose angular-momentum vectors are more strongly aligned with the rotation axis. Recovering the 3D orbit distribution requires the dynamical modeling we outline in § 6.
Detecting the P1 flattening requires azimuthal in addition to radial coverage: it is an intrinsically 2D signal that a purely radial profile would miss or dilute if the data under-sample one range of position angles. The combination of deep HST proper motions and wide-field MUSE LOS velocities in ωMEGACat provides the angular coverage this measurement needs.
The decomposition into P1, Im, and P2 may, in addition to the projected rotation (§ 4.1), partly explain the mild flattening of the cluster’s proper-motion dispersion field reported by M. Häberle et al. (2025), from a 2D Gaussian fit to the Voronoi-binned dispersion map. Their value is a dispersion ratio, not an ellipticity: on a nearly flat dispersion profile, a small azimuthal contrast gives a large (§ 3.2.2). The comparable quantity for P1 is (§ 4.1). P1 makes up about 30% of our chromosome-map sample (Appendix § 9), so if Im and P2 are round, the full sample would show a contrast of , comparable to theirs. Our bulk samples, which mix all three populations, give small values in the LOS (–; the smaller, central AMR sample is unconstrained) and, for the two larger samples, in the tangential proper motions (, consistent with zero), and only their radial proper-motion dispersion is more flattened (–).
Because Im and P2 share similar kinematics, we treat them as a single enriched component (P2/Im) throughout, returning to whether they represent one enrichment event or two in Section 5.7. Their radial distributions do differ: in the C. Clontz et al. (2026) sample, P2 is more centrally concentrated than Im (median radii of and ), whereas Im is distributed like P1 (median ; Figure 8).
These results constrain any viable formation scenario in three ways. First, P1 must have formed in, or later been brought into, a substantially flattened configuration: its kinematic ellipticity of requires either a birth environment with substantial angular momentum or geometric flattening, or a later event that induced the flattening dynamically, such as a merger or assembly event in which P1 fell in on a tangential orbit (§ 5.6). Second, P1 must rotate no faster than P2/Im within , and more slowly at . Third, P2/Im have a dispersion field consistent with spherical symmetry today, despite Cen’s long relaxation time, so they either formed round or became round early, while they were more compact than today, as their central concentration also suggests.
Helium abundances in the chromosome-map populations
Many previous studies of the spatial and kinematic properties of Cen have focused on the cluster’s multiple main sequences, splitting stars into He-normal and He-rich groups (A. Sollima et al. 2007; A. Bellini et al. 2009, 2018). The chromosome-map populations give us the closest equivalent of these populations on the RGB: at fixed [Fe/H], P2 is He-enhanced by – relative to P1, and Im only mildly (; C. Clontz et al. 2025). On the main sequence, the He-rich stars can be further split into the bMS, MSd, and MSa (A. Bellini et al. 2017). All three groups are He-enriched, but they differ in metallicity, from the bMS to the most metal-rich MSa. C. Clontz et al. (2026) map their subpopulations to the labels of A. Bellini et al. (2017): by number, the P2 stars are a mixture of the bMS (), MSd (), and MSa (), so P2 is therefore not a homogeneous population, though it contains only the He-rich groupings.
The spread of He within P2 is unlikely to affect our main results. The Im and P2 populations differ in He enhancement by a factor of – yet have similar kinematics within , and at larger radii the MSd- and bMS-like groups of A. Bellini et al. (2018) are both radially anisotropic, suggesting that the He-rich groups move similarly. We cannot exclude, however, that an outlier He-rich group with distinct kinematics is mixed into P2 and diluted by the other He-rich stars. On the MS, where these groups can be separated directly, such a group could be isolated (§ 6). The main signal we detect, the kinematic flattening of P1, is in the He-normal stars and is unaffected by the composition of P2.
Agreement with dynamical models and larger-radius studies
The signatures we observe in the inner regions are consistent with results at larger radii. At , the He-rich MS-II stars of A. Bellini et al. (2018), which correspond to our P2, are more radially anisotropic than the He-normal MS-I stars, with MS-I rotating faster; T. Ziliotto et al. (2026) (JWST NIRCam plus archival HST) find at – that their lower-stream stars (LS, analogous to our P1) are less radially anisotropic and show proper-motion skewness that suggests stronger rotation than the upper-stream stars (US, analogous to our P2/Im), with a higher tangential dispersion than the US stars at the same radial dispersion and a larger angular momentum dispersion. G. Pagnini et al. (2026) extend this comparison to – with APOGEE ([1] (S. R. Majewski et al. 2017)) data, where the Al-poor (P1-like) group is near-isotropic and the Al-rich (P2/Im-like) group is significantly radially anisotropic. Together, these studies show the primordial stars differing from the enriched stars outside in three ways: a larger angular-momentum dispersion at – (T. Ziliotto et al. 2026), weaker radial anisotropy out to large radii, where the two groups nonetheless rotate alike (G. Pagnini et al. 2026), and faster rotation at (A. Bellini et al. 2018). Within , where the cluster rotation curve flattens beyond (M. Häberle et al. 2025) and our fitted peak lies near the edge of the field (§ 4.1), P1 instead rotates more slowly than P2/Im (§ 4.1). P1 differs there in its orbital structure: a flattened dispersion field, consistent with angular-momentum vectors more strongly aligned with the rotation axis (§ 5.1). The differentially rotating equilibrium models of A. L. Varri & G. Bertin (2012) are flattened most strongly in their central regions and are nearly round in their outer parts, so the P1 flattening need not persist to larger radii. The faster rotation of P1 at can arise in two ways. P1 may have formed with more angular momentum than P2/Im. Alternatively, the initially compact enriched stars may have diffused outward (M. A. Tiongco et al. 2019; § 5.4). A P1 born in a rotating, flattened environment would account for both its inner flattening and its faster outer rotation, but the present data do not exclude the outward diffusion of the enriched stars.
Relaxation and Dynamical Evolution
Cospatial dynamical evolution can help explain the radial trend we observe. Two-body relaxation erases the differences between populations at different rates for different quantities, which is why some of the differences we observe survive while others may not. M. A. Tiongco et al. (2019) simulate an initially compact P2 population embedded in an extended P1, with the P2 rotating faster in the inner regions: two-body relaxation matches the rotation curves first in the inner regions, well before anisotropy differences are erased or spatial mixing completes. The P2 stars that diffuse outward rotate more slowly than P1 in the outermost regions, so P2 rotates faster over most of the cluster and P1 only in the outskirts, the pattern that our data and those of A. Bellini et al. (2018) show across the two radial ranges (§ 5.3). F. I. Aros et al. (2025) confirm in their -body models that anisotropy differences survive to (where is the number of elapsed half-mass relaxation times), long after concentration differences decay. Neither set of models is isolated, but both use a static tidal field: a circular orbit in a point-mass host (M. A. Tiongco et al. 2019) or a tidal-radius truncation (F. I. Aros et al. 2025). Neither includes the time-varying tides or disk shocks that Cen experiences on its eccentric orbit, which could heat the more extended P1 more strongly than the compact P2/Im.
Cen’s present-day half-mass relaxation time ( Gyr; H. Baumgardt & M. Hilker 2018; H. Baumgardt & E. Vasiliev 2021, updated online catalog) is long enough that one might expect even more pronounced differences than we observe. If, however, Cen expanded once its tidal field weakened on accretion into the MW halo (as M. Giersz et al. 2025 model for NGC 3201), the relaxation time would have been shorter in the past, and the populations may have mixed more than the present-day value implies.
Independently, P2’s helium enrichment (C. Clontz et al. 2025) makes its giants less massive than those of P1 ( for P1 vs. for P2; Z. Wang et al. 2025), because helium-rich stars evolve faster. The younger age of P2 works in the opposite direction, but this effect is much smaller. From equipartition alone (; M. Häberle et al. 2025), this mass difference predicts a dispersion difference of at most km s, several times smaller than the 3–4 km s difference between the major- and minor-axis dispersions of P1. A mass difference also changes the dispersion equally in all directions, so it cannot flatten the dispersion field, and equipartition cannot produce the kinematic differences we measure.
Comparison with Typical Globular Clusters
The kinematic differences we observe in Cen are similar to many of those found in typical globular clusters. There, the enriched stars are often, though not always, more centrally concentrated (e.g., E. Dalessandro et al. 2019; E. Leitinger et al. 2023), and they rotate faster than the primordial stars (E. Dalessandro et al. 2024) or at a similar rate (A. P. Milone et al. 2018; S. Martens et al. 2023; E. I. Leitinger et al. 2025). They become more radially anisotropic outside the core, while the primordial stars remain closer to isotropy (M. Libralato et al. 2023; E. Dalessandro et al. 2024; G. Cordoni et al. 2025). These differences are strongest in dynamically young clusters and fade with dynamical age (E. Dalessandro et al. 2024; G. Cordoni et al. 2025), and they match models in which the enriched stars form in the cluster core: such stars start centrally concentrated with faster rotation (V. Hénault-Brunet et al. 2015; E. Lacchin et al. 2022). NGC 2808, which also hosts strongly He-enhanced populations (G. Piotto et al. 2007), shows a similar pattern: its He-rich stars become radially anisotropic beyond through a lower tangential dispersion, while inside the populations are indistinguishable (A. Bellini et al. 2015; M. Griggio et al. 2025).
Our enriched stars largely follow this pattern: they are centrally concentrated, radially anisotropic outward (see also A. Bellini et al. 2018), and rotate at least as fast as P1 within , although at the primordial stars rotate faster (A. Bellini et al. 2018). P1 does not: no comparable flattening of the dispersion field has been reported for the primordial stars of typical clusters, whereas P1 has the most flattened dispersion field of any population in Cen. The flattened P1 component therefore has no counterpart in typical GCs, which suggests that it reflects Cen’s origin as a nucleus rather than the process that produced the enriched stars.
Formation Scenarios
Observational Constraints
Any formation scenario for Cen must reproduce six properties of its populations, constrained here and in earlier work:
Kinematics. Within , P1 is the flattened component and rotates – more slowly than P2/Im at , while P2 and Im are round and become radially anisotropic outward (§ 4.1).
Concentration. P2 is more centrally concentrated than P1 and Im, while Im is distributed like P1 (§ 5.1; Figure 8).
Ages. The population ages decrease from P1 ( Gyr) through Im ( Gyr) to P2 ( Gyr). P1 is older than P2 at every [Fe/H], and Im falls between P1 and P2 in mean age at every metallicity where it is present (C. Clontz et al. 2026). The most metal-rich stars of Cen formed at Gyr (C. Clontz et al. 2024).
Chemistry. [(C + N + O)/Fe] rises tightly with [Fe/H] with no offset among P1, Im, and P2, and [N/Fe] correlates tightly with age across all three populations. However, [Na/Fe], [O/Fe], and [Mg/Fe] follow a separate track for each population, and P2 is offset to higher [Ba/Fe] at fixed [Fe/H] (Z. Wang et al. 2026). P1 and P2 overlap widely in [Fe/H], and the large helium enhancement is already present in the most metal-poor P2 stars (C. Clontz et al. 2025).
Mass budget. Im and P2 together make up about two-thirds of the evolved stars within (C. Clontz et al. 2026).
Debris. In the halo substructures that S. O. Souza et al. (2026) associate with the host of Cen, P1-like stars dominate the outermost, earliest-stripped debris and follow the metallicity-spread sequence of dwarf galaxies, while the He–N–Al-rich P2-like stars are concentrated toward the inner regions and absent from the outskirts.
The kinematics can test whether the enriched stars formed dissipationally, from gas that cooled and settled in the cluster potential, or were assembled dissipationlessly from stars that had already formed. Stars that form from gas settled in a rotating potential inherit its angular momentum, and in simulations of second-generation formation the enriched stars form a faster rotating, flattened component (K. Bekki 2010; V. Hénault-Brunet et al. 2015; E. Lacchin et al. 2022).
The continuation of the paragraph appears in the next column: In M54 and the Milky Way nuclear star cluster, the metal-rich stars likewise have more rotational support than the metal-poor stars, which are possibly accreted (M. Alfaro-Cuello et al. [3]; N. Kacharov et al. 2022; A. Feldmeier-Krause et al. 2025). Stars assembled dissipationlessly, like these accreted components, are instead on radially biased orbits with less angular momentum (M. Alfaro-Cuello et al. [3]; N. Kacharov et al. 2022). The kinematics therefore argue against P2 and Im forming in a long-lived, cold disk, and favor either in-situ formation in which the disk is disrupted early (e.g., E. Lacchin et al. 2022), or assembly from clumps that arrived as stars. Figure 9 shows these two commonly proposed scenarios.
Figure 9. Schematic of the two formation scenarios for P2/Im discussed in this section, from P1’s initial flattened, rotating configuration (left) to the present day (right). Top row: under hierarchical assembly, P2/Im form as separate stellar clumps that spiral into P1’s potential via dynamical friction. Bottom row: under in-situ gas processing, P2/Im form directly from gas retained within P1’s own potential, without ever existing as a separate stellar system. Both pathways converge on the same present-day configuration (right): a primordial population with a flattened dispersion field and a centrally concentrated, spherical, radially anisotropic enriched population with similar rotation within . Both scenarios reproduce the kinematic constraints established in § 5.1, although in-situ gas processing does so only if the disk formed by the dissipating gas is erased early.
Scenario 1: Hierarchical Assembly
Under hierarchical assembly, P2/Im form as separate stellar clumps or star clusters that spiral to the center of the dwarf via dynamical friction (Figure 9, top row; see also K. Fahrion et al. 2021, 2022), as in models that build nuclear star clusters from inspiraling GCs (F. Antonini et al. [4]; A. Tsatsi et al. 2017; N. Neumayer et al. 2020). Hierarchical assembly can take several forms: Im and P2 both form in clumps detached from P1 and then fall in; only Im arrives as inspiraling GCs, and P2 forms afterward from the gas those clusters deliver to the center (A. C. Mason et al. 2026); sub-clusters merge while the cluster is still forming (e.g., N. Lahén et al. 2025); or a pre-formed, dense P2 system absorbs an infalling P1.
Separate infalling clumps would also deliver some primordial stars. These would be counted as P1 on the chromosome map, but would share the clump’s age and orbits, whereas P1 is older than P2 at every [Fe/H] and kinematically distinct from P2/Im. Im falls between P1 and P2 in age, an ordering that clumps forming and spiraling in one after another could reproduce, though only by coincidence. Individual clumps that condense from a larger chemically homogeneous parent cloud would inherit its composition, so the shared rise of [(C + N + O)/Fe] (A. F. Marino et al. 2012; Z. Wang et al. 2026) and the similar rise of the s-process elements ([La/Fe] in A. F. Marino et al. 2012, [Ba/Fe] in Z. Wang et al. 2026) with [Fe/H] can be explained by a common birth environment. One kinematic feature may point toward separate assembly for Im: its rotation axis lies from that of P2. In -body models of nuclear star cluster growth by cluster inspiral, a late infall changes the direction of the cluster’s angular momentum by – (A. Tsatsi et al. 2017), so a population delivered late could retain a distinct rotation axis.
If P1 instead fell onto a pre-formed, dense P2 system, the timeline is inverted. Because P2 is younger than P1 at every , P1 would then have formed elsewhere and fallen in only after P2 had formed. P1’s flattening would then be acquired from the orbit of the infall instead of inherited from its birth environment: in the inspiral models of A. Tsatsi et al. (2017), about two-thirds of the orbital angular momentum of the infalling clusters ends up in the nuclear cluster, which rotates and is flattened, although their models build the nuclear cluster from successive inspirals rather than a single infall. An infall that deposits enough orbital angular momentum to flatten P1 should also make it rotate faster, whereas P1 rotates more slowly than P2/Im at 200″. Im has no natural place in this version, however, and the chemical continuity across P1, Im, and P2 would have to arise in separate systems.
Scenario 2: In-situ Gas Processing
Under in-situ gas processing (Figure 9, bottom row), P2/Im instead form from gas retained within P1’s own potential, without ever existing as a separate system. Because every population then draws on the same enrichment sequence, this scenario reproduces the shared – and –age relations without fine-tuning. A single enrichment sequence is harder to reconcile with the separate Na, O, and Mg tracks of each population, with the large overlap in between P1 and P2, which requires strong dilution of the enriched gas, and with the large helium enhancement already present in the most metal-poor P2 stars. Variants in which Im forms by dilution (e.g., E. Dondoglio et al. 2026) are also in tension with the ages, since Im is older than P2.
The specific process that compresses and consumes the retained gas does not need to be pinned down. Compressive tides, as in galaxy mergers (F. Renaud et al. 2009, 2014), tidally driven gas inflow to the nucleus (K. Bekki & K. C. Freeman 2003), or a purely internal cooling flow (A. D’Ercole et al. 2008) all act primarily on gas rather than stars, and should leave the flattened dispersion field that P1 inherited untouched while forming P2/Im at the cluster center.
Gas that concentrates inside a rotating potential should settle into a disk (K. Bekki 2010), and -body models of second-generation disks in Cen-like clusters with up to of the cluster mass become nearly spherical within the Gyr of the simulations, while more massive disks remain flattened (A. Mastrobuono-Battisti & H. B. Perets 2013, 2016). The enriched populations together make up about two-thirds of the evolved stars within , so a second-generation disk of this size would be expected to survive to the present day. A round P2/Im therefore suggests either that its gas was dynamically hot when it formed stars, or that it formed at densities high enough to erase a disk quickly, as when infalling pristine gas disrupts the disk in the high-density models of E. Lacchin et al. (2022).
Debris, Mass Budget, and Timing
The debris of the disrupted progenitor argues against a late infall of P1 onto P2. P1 has counterparts across the host galaxy, whereas the enriched stars are found only near its center (S. O. Souza et al. 2026). This favors a P1 that formed as part of the host’s own stellar population, with its flattening inherited from the host’s inner, rotating body, over a P1 delivered by a single infalling cluster. Finding the enriched stars only near the center of the halo debris does not by itself show that they formed in the nucleus. Clusters with multiple populations also form away from the nucleus: NGC 5634, associated with Sagittarius, shows the Na–O and Mg–Al anticorrelations typical of GCs (E. Carretta et al. 2017), and clusters that sink to the center by dynamical friction would deliver their enriched stars there. Both scenarios can therefore place the enriched stars at the center, provided that in scenario 1 the clumps arrived before losing many of their enriched stars to the outer debris. Neither scenario is excluded, but neither accounts for the chemistry, the ages, and the mass budget together without further assumptions.
All of these scenarios share one problem, the mass budget: for a standard initial mass function, the polluters of a first generation cannot produce enough processed material to form enriched populations as numerous as those in Cen (A. Renzini et al. 2015; N. Bastian & C. Lardo 2018). We do not attempt to solve this problem here, but note that a galactic nucleus, as Cen likely was, eases it. Its deep potential retains the polluters’ ejecta (S. O. Souza et al. 2026) and draws in fresh gas from the host (K. Bekki & K. C. Freeman 2003). Ejecta that collect at the center, where P2 now lies, are retained most easily. The enriched gas can also be gathered from the whole central region of the progenitor, whose outer layers were later stripped (S. O. Souza et al. 2026). Its high stellar densities favor runaway stellar collisions (S. F. Portegies Zwart et al. 2004; N. Neumayer et al. 2020), which can build supermassive stars (P. A. Denissenkov & F. D. A. Hartwick 2014) that process a large fraction of the cluster mass (M. Gieles et al. 2018). Extremely massive stars that grow by accreting gas offer an alternative polluter (M. Gieles et al. 2025). Runaway collisions can leave an IMBH (S. F. Portegies Zwart et al. 2004; M. Gieles et al. 2018), which could have grown into the one at the center of Cen today (M. Häberle et al. 2024a).
Whichever scenario applies, the ages of the populations tie the enrichment to the infall of the progenitor. C. F. P. Laporte & M. D. A. Orkney (2026) date the pericentric passage of the GSE, the merger proposed to have brought Cen into the MW, to Gyr from a coeval burst of in-situ globular cluster formation, and the GSE’s most metal-rich clusters to Gyr. The most metal-rich stars of Cen formed at Gyr (C. Clontz et al. 2024), so if Cen was the GSE’s nucleus, its enriched populations may have formed during the GSE’s infall, and its star formation may have ceased once the GSE was disrupted (C. Clontz et al. 2024; C. F. P. Laporte & M. D. A. Orkney 2026; A. Mannu et al. 2026). In the simulations of C. F. P. Laporte & M. D. A. Orkney (2026), the infalling satellite’s central gas is compressed and forms stars on its final approach, before pericenter, which would supply an external trigger for the later bursts of enriched star formation.
One enrichment event or two?
The kinematics do not clearly separate Im from P2, but P2 is more centrally concentrated than Im, which is distributed like P1. Gas that was retained and settled at the center would form stars at least as concentrated as P2, so Im formed over a more extended region, before the enriched gas had settled as far inward, or was delivered from outside, as in the inspiraling-cluster model of A. C. Mason et al. (2026). The chemistry also separates them: the separate Na, O, and Mg tracks and the age ordering indicate that P1, Im, and P2 followed distinct enrichment tracks, whether or not they drew on a common reservoir, and that the continuity is limited to C+N+O and the N–age relation. Distinct tracks need not mean distinct gas sources, however. A single accretion event can bring the same gas into compressive tidal regions at successive pericenters, as in galaxy mergers (F. Renaud et al. 2009, 2014), and can drive repeated nuclear starbursts (K. Bekki & K. C. Freeman 2003). Kinematically, the data require no more than two components; the enriched component could have formed in one continuous episode or in successive bursts.
Limitations
There are a few limitations to the interpretation presented above.
First, the projected kinematics constrain the shape of the P1 dispersion field but not its intrinsic geometry. The axial ratio, vertical thickness, and inclination-corrected rotation amplitude remain degenerate in projection, and recovering them requires dedicated population-resolved dynamical modeling.
Second, the present-day central concentration and radial anisotropy of P2 reflect the outcome of P2’s formation, namely that it formed centrally concentrated, but not the mechanism for how it formed. The cluster’s dynamical age is itself uncertain, because Cen may have expanded once its tidal field weakened (§ 5.4), and its orbit may have been reshaped by resonances with the Galactic bar (A. M. Dillamore et al. 2026; A. Mannu et al. 2026), which may also strip stars along the orbit (A. Mannu et al. 2026). Both complicate the interpretation of P2’s present-day structure, although such external perturbations act mainly on the cluster’s outskirts and are unlikely to affect the kinematics we measure within .
Third, our population-resolved sample cannot determine which population hosts the counter-rotating component previously detected in the core of Cen (R. Pechetti et al. 2024; M. Häberle et al. 2025). The counter-rotation signal comes from the central –20\arcsec, covered mainly by MUSE’s Narrow Field Mode, where most stars lie below the magnitude limit of our population assignment (§ 3.1).
Fourth, our sample is limited to evolved stars (RGB and SGB), for which MUSE provides LOS velocities and metallicities. Extending the population decomposition to the lower main sequence, where the chromosome map pseudo-colors are available for all HST-detected stars without requiring MUSE spectra, would increase the number of kinematic tracers by an order of magnitude and allow tests of differential energy equipartition between populations, a step that the new JWST imaging of the cluster (M. Häberle et al., in prep.) will make possible.
Conclusions
We combined high-precision HST proper motions with MUSE line-of-sight velocities from the oMEGACat survey to analyze the 3D phase-space kinematics of the multiple stellar populations in Cen. By modeling the 2D dispersion, rotation, and proper-motion anisotropy fields, we reach the following conclusions about the cluster’s formation and assembly history:
Star-by-star abundance variations trace the dynamically distinct components. The chromosome-map division reveals a flattened P1 component ( in both HST proper motions and MUSE LOS velocities) and a P2/Im component that is consistent with in the LOS, and not significantly flattened in the proper motions. Divisions by mix stars from both components, which averages out these signatures, while the AMR sample is confined to the core, where the dispersion field is too flat to constrain any flattening. and AMR position likely reflect
the chemical evolution of the host galaxy (C. F. P. Laporte & M. D. A. Orkney 2026), while the chromosome map reflects enrichment within the cluster.
P1 has a distinct, flattened orbital structure. Within , P1 has a flattened dispersion field in both the LOS and the proper motions, consistent with orbits more strongly aligned with the rotation axis, and rotates more slowly than P2/Im at . At larger radii, P1 has a larger angular-momentum dispersion (T. Ziliotto et al. 2026), is less radially anisotropic ((A. Bellini et al. 2018; T. Ziliotto et al. 2026; G. Pagnini et al. 2026)), and rotates faster at (A. Bellini et al. 2018). The flattening suggests that P1 formed in a rotating, flattened environment. Its faster outer rotation may reflect more angular momentum at birth, or the outward diffusion of the initially compact enriched stars (M. A. Tiongco et al. 2019).
P2 formed more centrally concentrated than P1. P2/Im have dispersion fields consistent with spherical symmetry and are radially anisotropic beyond . P2 is more centrally concentrated than P1, while Im is distributed like P1. The central concentration of P2 suggests that P2 formed in a more compact configuration than P1 in each of the scenarios of § 5.6. The same compact configuration can also help to explain P2’s chemical enrichment: a deeper potential well retains a larger fraction of enriched ejecta.
Three extensions follow from these results. First, while the kinematic ellipticity () reported here is consistent with P1 forming a flattened structure, the precise 3D geometry (the intrinsic axial ratio, vertical thickness, and inclination-corrected rotation amplitude) remains unconstrained. Dedicated population-dynamical modeling of the 3D kinematics will be needed to recover the geometry of the P1 structure within the broader cluster. We will pursue this with population-resolved chemodynamical models in an upcoming paper (P. Smith et al., in prep.).
Second, these differences provide boundary conditions for formation modeling. Population-resolved -body simulations that track all three populations self-consistently over a Hubble time could constrain the cluster’s initial conditions and formation history. A forward-modeling approach that reproduces the present-day kinematics could constrain the formation of this nucleus and, by extension, the assembly history of its progenitor dwarf galaxy.
Finally, the current study is limited to evolved stars (RGB and SGB) where kinematic tracers are luminous enough for reliable PM and LOS measurements. Extending the sample to fainter MS stars that lack reliable MUSE spectra would increase the number of stars by an order of magnitude and allow us to search for effects like differential energy equipartition between the populations. HST and JWST pseudo-colors can already separate the populations on the MS (A. Bellini et al. 2017; M. Scalco et al. 2025; T. Ziliotto et al. 2026). New JWST imaging of the cluster (M. Häberle et al., in prep.) will allow us to extend this approach to the lower MS in a future paper.
Acknowledgments
P.J.S. is a fellow of, and acknowledges financial support from, the International Max Planck Research School for Astronomy and Cosmic Physics at the University of Heidelberg (IMPRS-HD).
It is a pleasure to thank the MPIA Galactic Nuclei and Milky Way groups for helpful discussions and feedback on this work. This work further benefited from discussions with the participants of the Galactic Nuclei Astrophysical Phenomena workshop at the Sexten Center for Astrophysics.
We thank Emanuele Dondoglio for kindly sharing with us the radial population fractions they measure in E. Dondoglio et al. (2026), which we plot in Figure 8.
A.F.K. acknowledges funding from the Austrian Science Fund (FWF) [grant DOI 10.55776/ESP542].
This research has made use of the Astrophysics Data System, funded by NASA under Cooperative Agreement 80NSSC21M00561. This work makes use of the colormaps in the CMasher package (E. van der Velden 2020; E. van der Velden et al. 2024). Software citation information was aggregated using The Software Citation Station (T. Wagg & F. S. Broekgaarden 2024; T. Wagg et al. 2025).
Anthropic’s Claude models were used to assist with code development, literature searches, and language editing of the manuscript. All generated content was reviewed and edited by the authors.
Facilities: VLT:Yepun(MUSE), HST(ACS, WFC3)
Software: astropy ( The Astropy Collaboration, et al. 2013, 2018, 2022; A. Collaboration 2026), ArviZ ({citep}R. Kumar et al. 2019; O. A. Martin et al. 2026), corner (D. Foreman-Mackey 2016), JAX (James Bradbury et al. 2018), jax-cosmo (J.-E. Campagne et al. 2023), matplotlib (J. D. Hunter 2007), numpy (C. R. Harris et al. 2020), numpyro (D. Phan et al. 2019; E. Bingham et al. 2019), pandas (W. McKinney 2010; The pandas development team 2026), python (G. Van Rossum & F. L. Drake 2009), …
Appendix
Sample Size per Subpopulation
Table A1 lists the number of stars in each subpopulation after each quality cut of § 2.4. The first row of each block counts the stars that meet the requirements of that division: valid chromosome-map indices and ; a fiducial age and [Fe/H] from C. Clontz et al. (2024); or with a MUSE metallicity. The cuts are cumulative: the third row gives the stars that pass both, and “Retained” the number as a fraction of the first row. For the AMR, the last row gives the stars used in the fits. We note that no population is preferentially removed by the cuts.
Chromosome Map: 1D Kinematic Profiles
For the 1D profiles, we use both discrete, unbinned methods and binned profiles, which serve as a check on each other. The discrete method adopts the same Plummer dispersion and Lynden-Bell rotation models used for the 2D fields (§ 3.2), but uses only the radial position of each star rather than its full 2D position. For the binned profiles, we use adaptive logarithmic radial bins (see, e.g., S. Kamann et al. 2018; M. Häberle et al. 2025), requiring a minimum of 150 stars per bin, and compute the dispersion and rotation within each bin, again implementing the likelihoods in JAX and sampling with NUTS. The binned profiles impose no functional form on the kinematics, allowing us to verify that the discrete models are appropriate for the data.
In one dimension, the chromosome-map populations already hint at the P1/P2 kinematic dichotomy but are more difficult to interpret than the 2D maps. For P1, the LOS dispersion profile is lower in the core and flatter with radius than those of 1m and P2, and the PM profiles show only modest differences between populations. These 1D trends are therefore best viewed as the radial projection of the stronger azimuthal structure seen in the 2D fields, rather than as independent evidence for a separate dynamical component. We compare each population with the other two populations combined over six radial bins. Across , , , , and the anisotropy, Im and P2 are consistent within uncertainties, except possibly in rotation axis; only P1 departs from the other populations, in and . The offset is in the sense of slower P1 rotation, as in the 2D fits (§ 4.1); its PM dispersions and anisotropy agree. The 1D profiles therefore do not distinguish the populations with the same clarity as the 2D maps.
Metallicity Bin Kinematics
The 1D kinematic profiles of the five quintiles show no significant differences in dispersion, rotation, or rotation-axis orientation. The 1D profiles are shown in Figure C1 and Figure C2. They are consistent with the 2D ellipticity fits (Figure 7), and indicate that the kinematics of the quintiles are consistent with each other within the uncertainties.
Age–Metallicity Relation Kinematics
Figure D1, Figure D2, Figure D3, and Figure D4 show the 1D profiles and 2D maps of the Tight and Diffuse AMR populations. These maps are harder to interpret than the chromosome-map results because of their radial coverage: the AMR requires F606W photometry, which confines the sample to the cluster core (§ 4.3). Across this range the dispersion varies only a little, and the few stars in the outermost bins dominate any measured gradient in the discrete fits. Within these limits, the Tight and Diffuse populations agree with each other and with the bulk in PM dispersion and anisotropy, and in LOS rotation amplitude and position angle. The one visible difference is the LOS dispersion of the Diffuse population, which stays nearly flat across the field while the Tight population follows the bulk decline (Figure D2). This difference rests on the sparsely populated outer bins and has no counterpart in the 2D ellipticity fits (Figure D5). Figure D5 shows the dispersion-field ellipticity posteriors of the Tight and Diffuse AMR populations.

Figure D1. 1D proper-motion kinematic profiles for the bulk AMR sample (black), Tight (green), and Diffuse (orange) populations. Panel layout as in Figure B2. The two populations show broadly similar PM dispersion and anisotropy profiles.
Figure D2. 1D LOS kinematic profiles for the bulk AMR sample (black), Tight (green), and Diffuse (orange) populations. Panel layout as in Figure B1. The Diffuse population shows a largely flat dispersion profile across the radial range probed, while the Tight population more closely tracks the overall cluster dispersion decline. No significant differences in rotation amplitude or position angle are detected.
Figure D3. 2D HST proper-motion kinematic maps for the AMR sample, Tight, and Diffuse populations (columns, left to right). Rows and axis lines as in Figure 4. Because the maps cover only the cluster core, where the dispersions vary little across the field, the shapes of these fields are weakly constrained; this includes the elongated Tight field (panel e).
Figure D4. 2D MUSE LOS rotation and velocity dispersion maps for the AMR sample, Tight, and Diffuse populations (columns, left to right). Rows and axis lines as in Figure 5. No significant kinematic flattening is detected; ellipticity posteriors are consistent with in both populations.
Rotation Posteriors
Figure E1 collects the LOS rotation posteriors of all three divisions. Because the rotation curve peaks at the edge of our field, the projected peak amplitude is degenerate with and its posteriors are broad. The rotation velocity on the kinematic major axis at , , lies inside the field and is computed for each posterior draw from the same fits, so it is better constrained; it is the rotation quantity we compare in the main text (§ 3.2).
References
- [1]Abdurro’uf, Accetta, K., Aerts, C., et al. 2022, The Astrophysical Journal Supplement Series, 259, 35, doi: 10.3847/1538-4365/ac4414
- [2]Akbaba, F., Plevne, O., Şahin, T., & Şentürk, S. A. 2026, Monthly Notices of the Royal Astronomical Society, 550, stag1308, doi: 10.1093/mnras/stag1308DOI
- [3]Alfaro-Cuello, M., Kacharov, N., Neumayer, N., et al. 2020, The Astrophysical Journal, 892, 20, doi: 10.3847/1538-4357/a77bb
- [4]Antonini, F., Capuzzo-Dolcetta, R., Mastrobuono-Battisti, A., & Merritt, D. 2012, The Astrophysical Journal, 750, 111, doi: 10.1088/0004-637X/750/2/111DOI
















