ALMA-IMF: Core Mass & Stellar IMF Insights

• ALMA-IMF is a homogeneous survey of 15 massive protoclusters that measures the core mass function (CMF) using matched physical resolution and dual-band imaging.
• The survey employs independent source-extraction methods and per-core temperature calibration via PPMAP to reveal both top-heavy and Salpeter-like CMF variations.
• By integrating continuum and spectral-line data, ALMA-IMF dissects kinematics, free-free emission, and feedback to shed light on protocluster dynamics and IMF origin.

ALMA-IMF is a large, homogeneous ALMA survey of 15 Milky Way massive protoclusters designed to measure the core mass function (CMF) in high-mass cluster-forming environments and to test how that CMF relates to the stellar initial mass function (IMF). Its defining features are matched physical resolution across the sample, dual-band continuum imaging at 1.3 mm and 3 mm, broad spectral-line coverage, and a programmatic emphasis on combining core demographics with kinematics, chemistry, gas inflow, feedback, and evolutionary state. Across the survey, ALMA-IMF has produced public continuum and line data products, core catalogs numbering from several hundred to about $10^3$ objects depending on definition and resolution, and a sequence of analyses that address whether massive protoclusters exhibit Salpeter-like or top-heavy CMFs, whether those CMFs evolve, and whether subsequent subfragmentation or accretion can reconcile them with the canonical IMF (Motte et al., 2021).

1. Survey definition, target selection, and evolutionary framework

ALMA-IMF was conceived to move CMF studies beyond nearby low-mass clouds and into the regime of clustered massive-star formation. The survey targets 15 protoclusters selected from the ATLASGAL-based catalog of the 200 most massive clumps in the Galactic plane, with the main selection criteria of integrated $870\,\mu\mathrm{m}$ flux larger than 25 Jy, a distance range of 2 to 5.5 kpc, exclusion of $355^\circ < l < 5^\circ$ because kinematic distances there are too uncertain, and representative sampling of evolutionary stages including both IR-quiet and IR-bright systems as well as W43 and W51 (Motte et al., 2021).

The final sample contains 15 massive protoclusters at distances of 2–5.5 kpc with a mean distance of 3.9 kpc. Their masses span roughly $2.5-33\times10^3\,M_\odot$, or, in the matched-resolution CMF study, a clump-mass range of 2500 to $32700\,M_\odot$ (Motte et al., 2021, Louvet et al., 2024). The survey mapped a total noncontiguous area of $\sim 53\ \mathrm{pc}^2$ at 1.3 mm and about $\sim 122\ \mathrm{pc}^2$ at 3 mm, while later science papers emphasize protocluster scales of order $1$–$3$ pc on a side and typical imaged regions around $2.5\times2.5\ \mathrm{pc}^2$ (Motte et al., 2021, Koley et al., 19 Jul 2025).

A central organizational principle is evolutionary classification. ALMA-IMF classifies protoclusters as Young, Intermediate, or Evolved based on the amount of dense gas in the cloud that has potentially been impacted by HII regions, using the 1.3 mm-to-3 mm flux ratio and free-free emission traced by H41$870\,\mu\mathrm{m}$0 (Motte et al., 2021). In that scheme, the sample contains 6 Young, 5 Intermediate, and 4 Evolved protoclusters. This framework underpins later comparisons of CMF shape, hot-core content, dense-gas tracers, outflow activity, and ionized-gas content across the sample.

Matched physical resolution is one of the survey’s core methodological commitments. Early program papers emphasized a near-uniform physical resolution of $870\,\mu\mathrm{m}$1 au at 1.3 mm and $870\,\mu\mathrm{m}$2 au at 3 mm, while the global CMF analysis later homogenized the data to a matched physical resolution of 2700 au to avoid resolution-driven biases in source size and recovered flux (Motte et al., 2021, Louvet et al., 2024). This homogeneous scaling is the basis for treating ALMA-IMF as a benchmark dataset for high-mass star formation across the Galaxy.

2. Observational architecture, imaging strategy, and released data products

ALMA-IMF observed each field in Band 6 at 1.3 mm and Band 3 at 3 mm. The continuum release from the 12 m array used products centered near 230.6 GHz in Band 6 and near 99.66 GHz in Band 3, with images created from a selection of the observed bandwidth using multi-term, multi-frequency synthesis imaging and deconvolution, self-calibration, and custom masking (Ginsburg et al., 2021). The program releases two continuum variants: cleanest, which selects portions of the spectrum unaffected by line emission, and bsens, which uses the full available bandwidth and therefore provides the best sensitivity at the cost of potential line contamination. Preliminary spectral-index analysis showed that these data clearly distinguish free-free emission from dust emission and in some cases identify optically thick emission sources (Ginsburg et al., 2021).

The first full spectral-line data release, DR1, combines two ALMA 12 m-array configurations and includes 12 spectral windows, with eight at 1.3 mm and four at 3 mm, providing $870\,\mu\mathrm{m}$3 GHz of bandwidth coverage per field (Cunningham et al., 2023). The line cubes were reprocessed in CASA because the archive pipeline alone could not always deliver stable deconvolution across the whole survey. The reprocessing introduced continuum-based start models, extensive QA reprocessing, multi-scale cleaning, primary-beam masking, careful threshold choices, and a JvM correction; the released cubes are continuum-subtracted, primary-beam-corrected, and JvM-corrected (Cunningham et al., 2023).

A later continuum-infrastructure paper combined ALMA-IMF interferometric images with single-dish surveys through feathering: six fields with MGPS90 at 3 mm and ten with BGPS at 1 mm, with six fields having combinations in both bands. Those combined maps enabled spectral-index separation of dust-dominated and free-free-dominated structures, and dendrogram-based structural analysis across clump and sub-clump scales (Díaz-González et al., 2023).

Another survey-wide infrastructure product derives free-free templates from H41$870\,\mu\mathrm{m}$4 recombination-line emission. These templates are matched to the ALMA continuum images and permit subtraction of free-free emission from Bands 3 and 6 continuum maps. The method complements spectral-index information and extrapolation from centimeter-wavelength maps, and it is central for more evolved protoclusters in which millimeter continuum is not purely dust emission (Galván-Madrid et al., 2024).

Product class	Main content	Programmatic role
12 m continuum release	Self-calibrated cleanest and bsens images in Bands 3 and 6	Core extraction, spectral-index analysis, dust/free-free separation
DR1 spectral cubes	12 spectral windows, eight at 1.3 mm and four at 3 mm	Kinematics, chemistry, inflow, outflows, hot cores
Feathered continuum maps	ALMA-IMF + MGPS90/BGPS combinations	Recovery of large-scale structure and dendrogram analysis
H41$870\,\mu\mathrm{m}$5-based templates	Spatially resolved free-free maps and pure-dust maps	Continuum correction and HII-region census

The observational design therefore couples high-resolution interferometric core studies to broader structural and spectroscopic context. This suggests that ALMA-IMF was intended not as a single CMF measurement but as a survey framework in which continuum, line cubes, and multi-scale products can be analyzed self-consistently.

3. Core catalogs, temperatures, masses, and the core mass function

The early ALMA-IMF catalog contained $870\,\mu\mathrm{m}$6 cores with masses of $870\,\mu\mathrm{m}$7 to $870\,\mu\mathrm{m}$8 at a typical deconvolved size of $870\,\mu\mathrm{m}$9 au, with no significant distance bias and with most cores above the thermal Bonnor–Ebert threshold (Motte et al., 2021). The large, matched-resolution CMF analysis later standardized the sample by smoothing the 1.3 mm and 3 mm maps to 2700 au and applying two independent source-extraction algorithms: getsf, a multi-scale extraction tool that separates sources, filaments, and background, and GExt2D, based on curvature or second-derivative detection followed by boundary identification and Gaussian fitting (Louvet et al., 2024).

At the matched 2700 au resolution, the survey identified about 680 gravitationally bound cores with getsf and about 1020 sources with GExt2D; roughly 80% of getsf sources are recovered by GExt2D (Louvet et al., 2024). The main CMF analysis adopted the getsf catalog because it is more conservative and better matched across the sample. The source list was then filtered to retain only likely thermal-dust cores: sources suspected of free-free contamination were removed using spectral indices derived from the 1.3 mm and 3 mm fluxes, sources with spectral indices indicative of strong free-free emission were excluded, and a first-order Bonnor–Ebert boundedness check was applied. After these cuts, the final core sample used for the CMF analysis contained 593 bound thermal-dust cores, and the global completeness-selected CMF fit used a subsample of 330 cores (Louvet et al., 2024).

A major methodological advance is the assignment of per-core temperatures using PPMAP rather than a single field temperature. In the global CMF paper, per-core dust temperatures typically span about 19.4 to 62.8 K, with an estimated uncertainty of about 5 K from the method, expanded to 25% to account for systematics; hot-core candidates were assigned higher temperatures guided by methyl formate emission and detailed modeling in the literature (Louvet et al., 2024). The later physical-calibration paper extended this approach to 883 securely classified cores and provided data-informed estimates of dust temperatures for 617 prestellar and 266 protostellar cores, with final masses spanning $355^\circ < l < 5^\circ$0 to $355^\circ < l < 5^\circ$1 (Motte et al., 2024).

The global completeness limit from injection-recovery tests is 1.64 $355^\circ < l < 5^\circ$2, meaning that above this mass the extraction is at least 90% complete; W51-E, which has a much higher completeness limit, was excluded from the global CMF fit (Louvet et al., 2024). Fitting the high-mass end with a maximum-likelihood estimator, the complete subsample of 330 cores above 1.64 $355^\circ < l < 5^\circ$3 yields

$355^\circ < l < 5^\circ$4

significantly flatter than the canonical Salpeter slope $355^\circ < l < 5^\circ$5 in the convention used there (Louvet et al., 2024). The paper states that this difference cannot reasonably be reconciled with Salpeter at the 2.4$355^\circ < l < 5^\circ$6 level, even after accounting for uncertainties in fluxes, temperatures, and dust opacity.

Regional analyses within ALMA-IMF show that the global result is not spatially uniform. In W43-MM2&MM3, a 205-core getsf catalog with masses from 0.1 to $355^\circ < l < 5^\circ$7 and 90% completeness down to $355^\circ < l < 5^\circ$8 yields a cumulative CMF slope

$355^\circ < l < 5^\circ$9

with a similar GExt2D result of $2.5-33\times10^3\,M_\odot$0, both top-heavy relative to the Salpeter IMF slope $2.5-33\times10^3\,M_\odot$1 (Pouteau et al., 2022). By contrast, the evolved W33-Main protocluster, after explicit correction for free-free and line contamination, gives

$2.5-33\times10^3\,M_\odot$2

slightly steeper than, but consistent with, the Salpeter index (Armante et al., 2024). One implication is that ALMA-IMF does not yield a single CMF morphology for all massive protoclusters; rather, it resolves both top-heavy and Salpeter-consistent cases within a common observational framework.

4. Evolutionary core populations, luminosity calibration, and prestellar lifetimes

ALMA-IMF progressively shifted from catalog construction to evolutionary census. In W43, a homogeneous core catalog combining W43-MM1 with W43-MM2&MM3 separates protostellar and prestellar candidates using molecular outflows as the main protostellar indicator and hot-core emission as a secondary indicator (Nony et al., 2023). In W43-MM1 there are 127 cores, and in W43-MM2&MM3 there are 205 cores. The resulting protostellar fractions are similar in the two regions, about 35%, but strongly mass dependent: $2.5-33\times10^3\,M_\odot$3 in the low-mass range $2.5-33\times10^3\,M_\odot$4–$2.5-33\times10^3\,M_\odot$5, around $2.5-33\times10^3\,M_\odot$6 in the 3.3–$2.5-33\times10^3\,M_\odot$7 bin, and up to $2.5-33\times10^3\,M_\odot$8 above $2.5-33\times10^3\,M_\odot$9 (Nony et al., 2023).

Protostellar cores in W43 are, on average, more massive and more compact than prestellar candidates. Above the common completeness threshold, the median properties are $32700\,M_\odot$0 and size $32700\,M_\odot$1 au for protostellar cores, versus $32700\,M_\odot$2 and size $32700\,M_\odot$3 au for prestellar cores (Nony et al., 2023). The prestellar high-mass CMF slope,

$32700\,M_\odot$4

is consistent with the Salpeter slope in the paper’s logarithmic convention, whereas the previously measured global W43 slope,

$32700\,M_\odot$5

is top-heavy. The paper therefore attributes the top-heavy global CMF primarily to the protostellar core population and interprets the result in terms of clump-fed models in which cores continue to grow in mass during the protostellar phase (Nony et al., 2023).

The survey-wide temperature and luminosity calibration of 883 cores further sharpened this evolutionary picture. For prestellar cores the 5th–95th percentile temperature range is 17–31 K, with median $32700\,M_\odot$6 K; for low-luminosity protostellar cores it is 26–51 K, with median $32700\,M_\odot$7 K; for luminous protostellar cores it is 38–92 K, with median $32700\,M_\odot$8 K and values up to 127 K (Motte et al., 2024). Protostellar luminosities span 20–80 000 $32700\,M_\odot$9, and the mass–luminosity diagram places many high-mass ALMA-IMF protostars at higher luminosity than predicted by declining-accretion tracks. The paper interprets this as evidence that protostars in massive protoclusters accrete from a larger, dynamically replenished reservoir than the core mass currently cataloged with ALMA (Motte et al., 2024).

ALMA-IMF XVII addresses the rarity and lifetime of high-mass prestellar cores in a statistically controlled way. Using CO and SiO outflows to classify the 141 most massive cores above $\sim 53\ \mathrm{pc}^2$0 in 14 protoclusters, it identifies 30 prestellar core candidates with $\sim 53\ \mathrm{pc}^2$1, including 12 with $\sim 53\ \mathrm{pc}^2$2, alongside 52 protostellar cores above $\sim 53\ \mathrm{pc}^2$3 (Valeille-Manet et al., 13 Feb 2025). With a baseline protostellar lifetime of 300 kyr, the derived prestellar lifetimes are 120 to 240 kyr for $\sim 53\ \mathrm{pc}^2$4 and 50 to 100 kyr for $\sim 53\ \mathrm{pc}^2$5, depending on mass-reservoir evolution scenarios (Valeille-Manet et al., 13 Feb 2025). These timescales significantly exceed the 4 to 15 kyr free-fall time of the cores, implying persistence over 10 to 30 free-fall times. The paper concludes that collapse is slowed by turbulence, magnetic fields, or rotation at or below the observed scale.

Taken together, these results recast ALMA-IMF core catalogs as an evolutionary inventory. They imply that core demographics in massive protoclusters are not static, and that the distinction between prestellar, protostellar, and hot-core phases is essential for interpreting any CMF–IMF mapping.

5. Dense-gas kinematics, turbulence, inflow, and outflow feedback

The survey’s spectroscopic component addresses the dynamical state of gas feeding core formation. DR1 uses DCN (3-2) as a dense-gas tracer associated with 595 continuum cores across the 15 protoclusters, extracting spectra in elliptical apertures matched to the continuum core sizes and fitting a hyperfine-structure model (Cunningham et al., 2023). DCN (3-2) is detected toward 357 of those cores, and 266 are well described by a single-component fit. The transition traces a diversity of morphologies and complex velocity structures, becoming more filamentary and widespread in evolved regions and more compact in young and intermediate protoclusters. Detection is markedly stage dependent: for single-component spectra the average detection rate is about 62% in evolved regions, compared with below 35% in young and intermediate regions (Cunningham et al., 2023).

At larger scales, ALMA-IMF XIX studies turbulence with C$\sim 53\ \mathrm{pc}^2$6O (2-1), smoothing the cubes to a common 2.5″ resolution and deriving local velocity dispersions through Gaussian decomposition (Koley et al., 19 Jul 2025). The resulting sonic Mach number distributions peak at $\sim 53\ \mathrm{pc}^2$7–7 and extend to $\sim 53\ \mathrm{pc}^2$8, with mean values typically $\sim 53\ \mathrm{pc}^2$9–9 and no clear trend with evolutionary stage. The non-thermal linewidth in dense DCN cores is, on average, half that of the surrounding C$\sim 122\ \mathrm{pc}^2$0O gas, suggesting that turbulence diminishes at smaller scales or dissipates at core boundaries (Koley et al., 19 Jul 2025). Dendrogram-based size-linewidth fits give power-law indices $\sim 122\ \mathrm{pc}^2$1 between 0.41 and 0.64, with a global $\sim 122\ \mathrm{pc}^2$2, steeper than the Kolmogorov incompressible expectation and consistent with compressible, shock-dominated turbulence.

Two target-specific N$\sim 122\ \mathrm{pc}^2$3H$\sim 122\ \mathrm{pc}^2$4 studies show how this turbulent medium organizes converging dense-gas flows. In G353.41, decomposition of the isolated N$\sim 122\ \mathrm{pc}^2$5H$\sim 122\ \mathrm{pc}^2$6 hyperfine component into up to 3 Gaussian velocity components reveals 9 converging V-shaped velocity gradients of order $\sim 122\ \mathrm{pc}^2$7, with average inflow timescale $\sim 122\ \mathrm{pc}^2$8 kyr, or about twice the free-fall time of nearby cores, and mass accretion rates in the range $\sim 122\ \mathrm{pc}^2$9 (Álvarez-Gutiérrez et al., 2024). The same paper finds good agreement between N$1$0H$1$1 velocities and previously reported DCN core velocities, indicating that cores are kinematically coupled to the dense gas in which they form.

In G351.77, N$1$2H$1$3 kinematics at $1$4 kau resolution reveal up to two velocity components, clear V-shapes associated with dense cores, a most prominent inflow with rate $1$5 and timescale $1$6 kyr, and a gas depletion time of $1$7 Myr (Sandoval-Garrido et al., 2024). Combined with the “Mother Filament,” these data motivate a picture of outside-in evolution in which the protocluster is continuously fed from larger scales.

Outflow feedback provides the complementary protostellar diagnostic. A blind SiO $1$8 search yields a catalog of 315 protostellar outflow candidates at $1$9 au resolution and 0.339 km s$3$0 velocity resolution (Towner et al., 2023). The median outflow properties are $3$1 in mass, $3$2 in momentum, $3$3 erg in kinetic energy, and a dynamical lifetime of 6000 years, corresponding to median rates of $3$4, $3$5, and $3$6 (Towner et al., 2023). Field-aggregated SiO outflow properties correlate with total mass in cores at the $3$7–$3$8 level, but show no correlation above $3$9 with clump mass, clump luminosity, or clump luminosity-to-mass ratio. The lack of correlation with clump $2.5\times2.5\ \mathrm{pc}^2$0 is explicitly described as inconsistent with models of protocluster formation in which all protostars start forming at the same time (Towner et al., 2023).

These kinematic studies establish that ALMA-IMF protoclusters are neither static nor uniformly turbulent. Instead, they are characterized by supersonic envelope gas, comparatively calmer dense cores, filamentary inflow, and distributed outflow feedback, all of which provide mechanisms for time-dependent modification of the CMF.

6. Hot cores, ionized gas, and interpretations of IMF origin

Chemistry and ionized-gas diagnostics are essential because they identify embedded high-mass protostars, constrain temperatures, and expose contamination of dust-continuum fluxes. In W43-MM1, ALMA-IMF IV identified at least 1 less massive and 7 massive hot cores using compact continuum sources, strong complex organic molecule emission, line-rich spectra, and elevated line-to-continuum ratios (Brouillet et al., 2022). CH$2.5\times2.5\ \mathrm{pc}^2$1CN excitation temperatures are 120–160 K for most hot cores, with uncertainties of about $2.5\times2.5\ \mathrm{pc}^2$2 K, while CH$2.5\times2.5\ \mathrm{pc}^2$3CCH gives lower temperatures of 50–90 K and preferentially traces the envelope. After scaling spectra to CH$2.5\times2.5\ \mathrm{pc}^2$4OCHO line intensities, the relative intensities of many species agree within a factor of 2–3, indicating broadly similar complex organic molecule compositions across hot cores spanning roughly an order of magnitude in core mass (Brouillet et al., 2022).

At the survey level, ALMA-IMF XI built a census of 76 CH$2.5\times2.5\ \mathrm{pc}^2$5OCHO-selected hot core candidates across all 15 protoclusters, of which 56 are new detections (Bonfand et al., 2024). Their continuum-derived masses span about 0.2 to $2.5\times2.5\ \mathrm{pc}^2$6, with median about $2.5\times2.5\ \mathrm{pc}^2$7; deconvolved FWHM sizes span about 990 to 13,400 au, with a typical size near 2300 au. About 30% of the sample have core masses above $2.5\times2.5\ \mathrm{pc}^2$8 and correspond well to archetypical hot cores, while the lower-mass CH$2.5\times2.5\ \mathrm{pc}^2$9OCHO sources may reflect shocks, accretion-related sputtering, or more evolved objects beyond the classical hot-core stage (Bonfand et al., 2024). The fraction of hot core candidates rises strongly with core mass, and the paper summarizes the massive-core result conservatively as “massive cores spend at least $870\,\mu\mathrm{m}$00 yr in the hot core phase.”

Ionized gas enters the ALMA-IMF framework through H41$870\,\mu\mathrm{m}$01-derived free-free templates and an HII-region census. Up to 34 HII regions are identified across the protoclusters where H41$870\,\mu\mathrm{m}$02 is detected, with total hydrogen ionizing-photon rates increasing from $870\,\mu\mathrm{m}$03 to $870\,\mu\mathrm{m}$04 along the evolutionary sequence (Galván-Madrid et al., 2024). The paper finds

$870\,\mu\mathrm{m}$05

and argues that smaller ultracompact HII regions are not necessarily less dynamically evolved versions of larger ones, but rather are ionized by less massive stars. Relative helium abundances measured from He41$870\,\mu\mathrm{m}$06 are mostly consistent with the Galactic interstellar medium, and a search for H41$870\,\mu\mathrm{m}$07 maser amplification yields a negative result (Galván-Madrid et al., 2024).

The broader interpretive significance of ALMA-IMF concerns IMF universality. The global CMF result of $870\,\mu\mathrm{m}$08, flatter than the Salpeter $870\,\mu\mathrm{m}$09, implies under self-similar CMF-to-IMF mapping that the 15 high-mass protoclusters would generate atypical IMFs (Louvet et al., 2024). Yet the program also identifies stage dependence and internal evolution. In W43-MM2&MM3, six subregions show CMFs ranging from Salpeter-like in quiescent zones to top-heavy in burst and post-burst zones, with flatter second tails in the high-column-density PDF associated with the top-heavy CMFs (Pouteau et al., 2022). The paper proposes that the CMF may evolve from Salpeter to top-heavy from the quiescent to the burst phase and raises the possibility that it might revert again to Salpeter near the end of the star-formation stage.

A more recent fragmentation analysis of W43-MM1 extends this debate below the core scale. Using five 3 mm images from 14 kau to 270 au and the FAMILY graph-theory tool, ALMA-IMF XXII measures a small fractality index $870\,\mu\mathrm{m}$10, a sibling mass partition of about $870\,\mu\mathrm{m}$11, and a core formation efficiency from 2400 au to 200 au of $870\,\mu\mathrm{m}$12 (Motte et al., 16 Apr 2026). The resulting fragment mass function remains top-heavy, with

$870\,\mu\mathrm{m}$13

after starting from a top-heavy core-scale slope $870\,\mu\mathrm{m}$14. The paper concludes that core subfragmentation plays only a minimal role in the IMF origin in W43-MM1. This suggests that, within the ALMA-IMF framework, later subfragmentation alone is not an obvious mechanism for restoring a canonical IMF once a top-heavy CMF is established.

ALMA-IMF therefore occupies a distinctive position in contemporary star-formation research. It provides a homogeneous observational basis for comparing core demographics across 15 massive protoclusters, while also showing that CMF shape, kinematics, chemistry, and feedback are coupled to evolutionary state and environment. The accumulated program results support neither a purely static one-to-one core-to-star mapping nor a single universal route from CMF to IMF; instead, they describe a massive-protocluster regime in which top-heavy core populations, clump-fed growth, regulated collapse, and early feedback must all be considered together.