Jouni Kainulainen

Context . Modern telescopes generate increasingly large and diverse datasets, often consisting of complex and morphologically rich structures. To efficiently explore such data requires automated methods that can extract and organize physically meaningful information, ideally without the need for extensive manual interaction. Aims . Our aim is to provide a user-friendly implementation of a self-supervised machine learning framework to explore morphological properties of large datasets, based on the Bootstrap Your Own Latents (BYOL) method. By enabling the generation of meaningful image embeddings without manually labelled data, the framework will enable key tasks such as clustering, anomaly detection, and similarity-based exploration. Methods . We present astromorph , a Python package that implements the BYOL method in a way tailored for astronomical imaging. In contrast to existing BYOL implementations, astromorph accommodates data of varying dimensions and resolutions, including both single-channel FITS images and multi-channel spectral cubes. The package is built with usability in mind, and offers streamlined pipeline scripts for ease of use as well as deeper customization options via PyTorch-based classes. Results . To demonstrate the utility of astromorph , we apply it in two contrasting science cases representing different astronomical domains: images of protoplanetary disks observed with the Atacama Large Millimeter/submillimeter Array (ALMA), and infrared dark clouds observed with Spitzer and Herschel . In both cases we demonstrate how astromorph produces scientifically meaningful embeddings that capture morphological differences and similarities across large samples. Conclusions . astromorph enables users to apply a robust, label-free approach for uncovering morphological patterns in astronomical datasets. The successful application to two markedly different datasets suggests that the pipeline is broadly applicable across a wide range of imaging-rich astronomical contexts, providing a user-friendly tool for advancing discovery in observational astronomy.

Context. Volume density is a key physical quantity that controls the evolution of the interstellar medium (ISM) and star formation, but it cannot be accessed directly by observations of molecular clouds. Aims. Our aim is to estimate the volume density distribution in nearby molecular clouds in order to measure the relation between column and volume densities and to determine their roles as predictors of star formation. Methods. We developed an inverse modelling method to estimate the volume density distributions of molecular clouds. We applied this method to 24 nearby molecular clouds for which column densities had been derived using Herschel observations and for which star formation efficiencies (SFE) had been derived using observations with the Spitzer space telescope. We then compared the relationships of several column-density-based and volume-density-based descriptors of dense gas with the SFEs of the clouds. Results. We derived volume density distributions for 24 nearby molecular clouds, which represents the most complete sample of such distributions to date. The relationship between column densities and peak volume densities in these clouds is a piecewise power law relation that changes its slope at a column density of 5–10 × 10²² H₂ cm⁻³. We interpret this as a signature of hierarchical fragmentation in the dense ISM. We find that the volume-density-based dense gas fraction is the best predictor of star formation in the clouds, and in particular, it is as anticipated a better predictor than the column-density-based dense gas fraction. We also derived a volume density threshold density for star formation of 2 × 10⁴ H₂ cm⁻³.

Context. Identified as parsec-size, gas clumps at the junction of multiple filaments, hub-filament systems (HFS) play a crucial role during the formation of young clusters and high-mass stars. These HFS still appear to be detached from most galactic filaments when compared in the mass–length (M–L) phase space. Aims. We aim to characterize the early evolution of HFS as part of the filamentary description of the interstellar medium (ISM). Methods. Combining previous scaling relations with new analytic calculations, we created a toy model to explore the different physical regimes described by the M–L diagram. Despite its simplicity, our model accurately reproduces several observational properties reported for filaments and HFS, such as their expected typical aspect ratio (A), mean surface density (Σ), and gas accretion rate (ṁ). Moreover, this model naturally explains the different mass and length regimes populated by filaments and HFS, respectively. Results. Our model predicts a dichotomy between filamentary (A ≥ 3) and spheroidal (A < 3) structures connected to the relative importance of their fragmentation, accretion, and collapse timescales. Individual filaments with low accretion rates are dominated by an efficient internal fragmentation. In contrast, the formation of compact HFS at the intersection of filaments triggers a geometric phase-transition, leading to the gravitational collapse of these structures at parsec-scales in ~1–2 Myr. In addition, this process also induces higher accretion rates.

Context. The kinematics of molecular gas are crucial for setting the stage for star formation. One key question related to the kinematic properties of gas is how they depend on the spatial scale. Aims. We aim to describe the CO spectra, velocity dispersions, and especially the linewidth-size relation of molecular gas from cloud (parsec) scales to kiloparsec scales in a complete region within the Milky Way disk. Methods. We used the census of molecular clouds within 2 kpc from our earlier work, together with CO emission data for them from the literature. We studied the kinematics and the Larson relations for the sample of individual clouds. We also mimicked a face-on view of the Milky Way and analysed the kinematics of the clouds within apertures of 0.25–2 kpc in size. In this way, we describe the scale-dependence of the CO gas kinematics and Larson’s relations. Results. We describe the spectra of CO gas at cloud scales and in apertures between 0.25 and 2 kpc in our survey area. The spectra within the apertures are relatively symmetric, but show non-Gaussian high-velocity wings. At cloud scales, our sample shows a linewidth-size relation σ_v = 1.5 · R^0.3±0.1 with a large scatter. The mass-size relation in the sample of clouds is M_CO = 794 · R^1.5±0.5. The relations are also present for the apertures at kiloparsec-scales. The best-fit linewidth-size relation for the apertures is σ_v = 0.5 · R^0.35±0.01, and the best-fit mass-size relation is M_CO = 229 · R^1.4±0.1. A suggestive dependence on Galactic environment is seen. Apertures closer to the Galactic centre and the Sagittarius spiral arm have slightly higher velocity dispersions. We explore the possible effect of a diffuse component in the survey area and find that such a component would widen the CO spectra and could flatten the linewidth-size relation. Understanding the nature of the possible diffuse CO component and its effects on observations is crucial for connecting Galactic and extragalactic data.

The new Gaia data release (EDR3) with improved astrometry has opened a new era in studying our Milky Way in fine detail. We use Gaia EDR3 astrometry together with 2MASS and WISE photometry to study two of the most massive molecular clouds in the solar vicinity: Orion A and California. Despite having remarkable similarities in the plane of the sky in terms of shape, size, and extinction, California has an order of magnitude lower star formation efficiency. We use our state-of-the-art dust mapping technique to derive the detailed three-dimensional (3D) structure of the two clouds, taking into account both distance and extinction uncertainties, and a full 3D spatial correlation between neighboring points. We discover that, despite the apparent filamentary structure in the plane of the sky, California is a flat 120 pc-long sheet extending from 410 to 530 pc. We show that not only Orion A and California differ substantially in their 3D shapes, but also Orion A has considerably higher density substructures in 3D than California. This result presents a compelling reason why the two clouds have different star formation activities. We also demonstrate how the viewing angle of California can substantially change the cloud’s position in the Kennicutt–Schmidt relation. This underlines the importance of 3D information in interpreting star formation relations and challenges studies that rely solely on the column density thresholds to determine star formation activities in molecular clouds. Finally, we provide accurate distance estimates to multiple lines of sight toward various parts of the two clouds.

The Gaia data give us an unprecedented view to the three-dimensional (3D) structure of molecular clouds in the solar neighbourhood. We study how the projected areas and masses of clouds, and consequently the Kennicutt-Schmidt (KS) relation, depend on the viewing angle. We derive the probability distributions of the projected areas and masses for nine clouds within 400 pc of the Sun using 3D dust distribution data from the literature. We find that the viewing angle can have a dramatic effect on the observed areas and masses of individual clouds. The joint probability distributions of the areas and masses are strongly correlated, relatively flat, and can show multiple peaks. The typical ranges and 50% quartiles of the distributions are roughly 100–200% and 20–80% of the median value, respectively, making viewing angle effects important for all individual clouds. The threshold value used to define the cloud areas is also important; our analysis suggests that the clouds become more anisotropic for smaller thresholds (larger scales). On average, the areas and masses of the plane-of-the-sky and face-on projections agree, albeit with a large scatter. This suggests that sample averages of areas and masses are relatively free of viewing angle effects, which is important to facilitate comparisons of extragalactic and Galactic data. Ultimately, our results demonstrate that a cloud’s location in the KS relation is affected by the viewing angle in a non-trivial manner. However, the KS relation of our sample as a whole is not strongly affected by these effects, because the covariance of the areas and masses causes the observed mean column density to remain relatively constant.

Context. Describing how the properties of the interstellar medium are combined across various size scales is crucial for understanding star formation scaling laws and connecting Galactic and extragalactic data of molecular clouds. Aims. We describe how the statistical structure of the clouds and its connection to star formation changes from sub-parsec to kiloparsec scales in a complete region within the Milky Way disk. Methods. We built a census of molecular clouds within 2 kpc from the Sun using data from the literature. We examined the dust-based column density probability distributions (N-PDFs) of the clouds and their relation to star formation as traced by young stellar objects (YSOs). We then examined our survey region from the outside, within apertures of varying sizes, and describe how the N-PDFs and their relation to star formation changes with the size scale. Results. We present a census of the molecular clouds within 2 kpc distance, including 72 clouds and YSO counts for 44 of them. The N-PDFs of the clouds are not well described by any single simple model; use of any single model may bias the interpretation of the N-PDFs. The top-heaviness of the N-PDFs correlates with star formation activity, and the correlation changes with Galactic environment (spiral- and inter-arm regions). We find that the density contrast of clouds may be more intimately linked to star formation than the dense gas mass fraction. The aperture-averaged N-PDFs vary with the size scale and are more top-heavy for larger apertures. The top-heaviness of the aperture N-PDFs correlates with star formation activity up to roughly 0.5 kpc, depending on the environment. Our results suggest that the relations between cloud structure and star formation are environment specific and best captured by relative quantities (e.g. the density contrast). Finally, we show that the density structures of individual clouds give rise to a kiloparsec-scale Kennicutt-Schmidt relation as a combination of sampling effects and blending of different galactic environments.

Context. High-mass stars and star clusters commonly form within hub-filament systems. Monoceros R2 (hereafter Mon R2), at a distance of 830 pc, harbors one of the closest of these systems, making it an excellent target for case studies. Aims. We investigate the morphology, stability and dynamical properties of the Mon R2 hub-filament system. Methods. We employed observations of the ¹³CO and C¹⁸O 1 →0 and 2 →1 lines obtained with the IRAM-30 m telescope. We also used H₂ column density maps derived from Herschel dust emission observations. Results. We identified the filamentary network in Mon R2 with the DisPerSE algorithm and characterized the individual filaments as either main (converging into the hub) or secondary (converging to a main filament). The main filaments have line masses of 30–100 M_⊙ pc⁻¹ and show signs of fragmentation, while the secondary filaments have line masses of 12–60 M_⊙ pc⁻¹ and show fragmentation only sporadically. In the context of Ostriker’s hydrostatic filament model, the main filaments are thermally supercritical. If non-thermal motions are included, most of them are transcritical. Most of the secondary filaments are roughly transcritical regardless of whether non-thermal motions are included or not. From the morphology and kinematics of the main filaments, we estimate a mass accretion rate of 10⁻⁴–10⁻³ M_⊙ yr⁻¹ into the central hub. The secondary filaments accrete into the main filaments at a rate of 0.1–0.4 × 10⁻⁴ M_⊙ yr⁻¹. The main filaments extend into the central hub. Their velocity gradients increase toward the hub, suggesting acceleration of the gas. We estimate that with the observed infall velocity, the mass-doubling time of the hub is ~2.5 Myr, ten times longer than the free-fall time, suggesting a dynamically old region. These timescales are comparable with the chemical age of the HII region. Inside the hub, the main filaments show a ring- or a spiral-like morphology that exhibits rotation and infall motions. One possible explanation for the morphology is that gas is falling into the central cluster following a spiral-like pattern.

We introduce the histogram of oriented gradients (HOG), a tool developed for machine vision that we propose as a new metric for the systematic characterization of spectral line observations of atomic and molecular gas and the study of molecular cloud formation models. In essence, the HOG technique takes as input extended spectral-line observations from two tracers and provides an estimate of their spatial correlation across velocity channels. We characterized HOG using synthetic observations of HIand¹³CO (J= 1 → 0) emission from numerical simulations of magnetohydrodynamic (MHD) turbulence leading to the formation of molecular gas after the collision of two atomic clouds. We found a significant spatial correlation between the two tracers in velocity channels wherev_HI≈v_13CO, almost independent of the orientation of the collision with respect to the line of sight. Subsequently, we used HOG to investigate the spatial correlation of the HI, from The HI/OH/recombination line survey of the inner Milky Way (THOR), and the¹³CO (J= 1 → 0) emission from the Galactic Ring Survey (GRS), toward the portion of the Galactic plane 33°.75 ≤l≤ 35°.25 and |b| ≤ 1°.25. We found a significant spatial correlation between the two tracers in extended portions of the studied region. Although some of the regions with high spatial correlation are associated with HIself-absorption (HISA) features, suggesting that it is produced by the cold atomic gas, the correlation is not exclusive to this kind of region. The HOG results derived for the observational data indicate significant differences between individual regions: some show spatial correlation in channels aroundv_HI≈v_13COwhile others present spatial correlations in velocity channels separated by a few kilometers per second. We associate these velocity offsets to the effect of feedback and to the presence of physical conditions that are not included in the atomic-cloud-collision simulations, such as more general magnetic field configurations, shear, and global gas infall.