WorksPhysics and Astronomy
Probing the Turbulent Magnetised Interstellar Medium of the Small Magellanic Cloud through Depolarisation with POSSUM Observations
Abstract
The interstellar medium (ISM) of the Small Magellanic Cloud (SMC) provides a unique laboratory for studying magnetised gas in a low-metallicity, tidally disturbed environment. We investigate this magnetised turbulent medium using broad-band polarimetric data from the Australian Square Kilometre Array Pathfinder (ASKAP)'s Polarisation Sky Survey of the Universe's Magnetism (POSSUM), leveraging a high-density Rotation Measure (RM) grid comprising 951 RMs ( RMs per square degree) to probe the magneto-ionic structure of the SMC. We statistically compare the number densities of polarised and non-polarised non-thermal background sources on a per-square-degree basis, revealing spatially coherent depolarisation signatures that correlate with enhanced plane-of-sky RM fluctuations. This is consistent with a turbulent cascade from large-scale structures. Analysed through RM synthesis and Stokes -fitting, we attribute the depolarisation to a turbulent magneto-ionic medium within the telescope's beam (). Faraday complexity is detected in 24% of the RM sources; however, no coherent spatial structure is observed in the complexity distribution, and we attribute this primarily to internal source structure or an extended halo surrounding the SMC rather than to its main body. Furthermore, as previously identified in the Large Magellanic Cloud, depolarisation in the SMC increases with emission measure; however, the underlying Faraday dispersion is 6 times weaker. With the thermal electron density smooth on the scales of the field fluctuations, this depolarisation is likely driven by magnetic field fluctuations.
Introduction
The detailed structure of magnetic fields in the interstellar medium (ISM) of galaxies has largely remained elusive observationally due to the sparsity of high-sensitivity data. It is well established that magnetic fields thread through the ionised, atomic, and molecular phases of the ISM ([67, 71] (Ferrière 2020; Seta & McClure-Griffiths 2025)) and influence the structure and flow of gas (Heiles & Haverkorn 2012; Gent et al. 2013; Clark et al. 2014; Inoue & Inutsuka 2016; Evirgen et al. 2016; Villagran & Gazol 2018; Gazol & Villagran 2018; Bracco et al. 2020; Seta & Federath 2022; Campbell et al. 2022; Jung et al. 2023). Conversely, large-scale motions and turbulence within the ISM modify the properties of magnetic fields (Brandenburg & Subramanian 2005; Rincon 2019; Seta et al. 2020). The complex dynamical interplay between the gas, turbulence, and magnetic fields generates ISM structures shaped by both small-scale processes, such as star formation and supernova explosions (McClure-Griffiths et al. 2006; Lyutikov 2006; Hidalgo-Pineda et al. 2023; Jung et al. 2024; Granda-Muñoz et al. 2025), and large-scale processes, including galactic-scale shear and outflows, and dynamical interactions in merging systems (Shukurov & Subramanian 2021; Brandenburg & Ntormousi 2023).
Studies of magnetised structures in a variety of astrophysical scenarios have advanced rapidly with the increasing capabilities of modern radio telescopes. The Polarisation Sky Survey of the Universe’s Magnetism (POSSUM; Gaensler et al. 2025) survey, underway with the Australian Square Kilometre Array Pathfinder (ASKAP; Hotan et al. 2021), is using Faraday rotation and polarised synchrotron emission to measure Galactic and extragalactic magnetic fields with an unprecedented combination of sensitivity, bandwidth, and sky area.
In this paper, we exploit POSSUM Pilot II (800–1088 MHz) observations to study the magnetic structure of the ISM in the Small Magellanic Cloud (SMC). The SMC is a tidally disrupted dwarf satellite galaxy of the Milky Way (MW), located 62.4 ± 0.8 kpc away (Graczyk et al. 2020). It is thought to be on its first passage around the MW, in a tidal interaction with its companion, the Large Magellanic Cloud (LMC). This interaction has given rise to several features, including the Magellanic Bridge (Nidever et al. 2013; Zivick et al. 2018), the Leading Arm (Putman et al. 1998; Nidever et al. 2008), and the Magellanic Stream (Mathewson et al. 1974; D’Onghia & Fox 2016). The SMC itself comprises two primary regions: the Bar, which contains the majority of its gaseous and stellar mass, and the Wing, a tidal feature extending towards the Magellanic Bridge that connects the SMC and LMC. The SMC is characterised by low gas metallicity and low star formation rates (D’Onghia & Fox 2016). Due to its proximity, the SMC is an excellent target for studying magnetic fields in an interacting dwarf galaxy ([71] (Haynes et al. 1986; Mathewson & Ford 1970; Mao et al. 2008; Livingston et al. 2022; Seta et al. 2023; Jung et al. 2024)). In this work, we primarily use the Faraday rotation effect to probe the magnetised ISM of the SMC.
Faraday rotation refers to the change in the observed angle of linear polarisation () from its intrinsic value at the source () according to the relation: , where the degree of rotation is quantified by the Rotation Measure (RM) or Faraday depth ()2 (Burn 1966; Brentjens & de Bruyn 2005; Klein & Fletcher 2015) expressed as:
where is the thermal electron density, is the magnetic field strength along the line of sight (LOS), is the differential path length element along the LOS, and represents the path length through the Faraday-rotating component; the sign of the represents the mean direction of the LOS magnetic field, positive towards the observer and negative away (see Rybicki & Lightman 1979; Ferrière et al. 2021). The polarised intensity (PI) of the source, expressed as a complex vector in the Stokes plane, is , where the linear polarisation angle is given by . When measured across a broad band, the polarised intensity enables characterisation of the magneto-ionic medium through which the emission has propagated.
However, variations in and/or on spatial scales smaller than the beam can introduce heterogeneity in the Faraday rotation experienced by the polarised rays of light within the beam, leading to depolarisation. When these rays are convolved on the plane of sky (POS), the heterogeneity across their LOS Faraday rotation produces wavelength-dependent depolarisation (beam depolarisation). In the limit of infinitesimal angular resolution, all individual polarised rays would be fully resolved, and beam depolarisation would therefore vanish (for background sources). This depolarisation mechanism induces a random walk in the polarisation vector due to the stochastic Faraday rotation and is wavelength dependent (Burn 1966; Sokoloff et al. 1998; Klein & Fletcher 2015). The standard deviation of Faraday depths realised during this random walk is quantified by the Faraday dispersion:
where is the correlation length scale of the Faraday-rotating plasma (Klein & Fletcher 2015; Orǹ et al. 2019; Seta & Feder-rath 2021). Under sufficiently strong Faraday dispersion, the polarised intensity of background sources may be suppressed below detectability (Tribble 1991; Sokoloff et al. 1998; Stil & Taylor 2007; Shneider et al. 2014; Machida et al. 2018). While this leads to the loss of direct Faraday rotation information, the spatial distribution of these non-detections is itself a measure of Faraday dispersion regions, with fewer polarised sources indicating stronger Faraday dispersion, assuming the sources are intrinsically polarised, providing a powerful diagnostic of the turbulent magneto-ionic medium.
In this work, we present 951 RMs ( RMs per square degree) across the field of the SMC, delivered by the pilot observations of the POSSUM survey, representing an approximately fivefold increase compared to previous studies (Mao et al. 2008; Livingston et al. 2022). This dense “RM grid” enables a detailed assessment of the SMC’s magneto-ionic structure through an analysis of depolarisation. At the SMC’s distance of 62.4 kpc (Graczyk et al. 2020), our synthesised beam of 21″ corresponds to a spatial resolution of pc.
This paper is organised as follows. § 2 describes the POSSUM observations, data reduction procedures, and source-finding methodology. In § 3, we outline the methods used to assess depolarisation and Faraday complexity. § 4 presents the results, while § 5 provides further analysis and discusses these findings in the context of the prior observations. We conclude in § 6.
Observational data
POSSUM Observations of the SMC
Spectro-polarimetric observations were made on the 7th August 2022 with a 10-hour integration time. The observations followed POSSUM’s Band 1 specifications covering 288 MHz spanning 800–1088 MHz with a 1 MHz sampling, over approximately 40 square degrees of sky. We achieved a full-band sensitivity of 27 Jy beam in Stokes and 17 Jy beam in Stokes and , respectively. Upon inspection of our Stokes multi-frequency synthesis (MFS) image, substantial diffuse continuum emission from within the SMC was apparent. To prevent the obscuration of background sources by the SMC during our source-finding procedures, we subtracted its diffuse Stokes emission from the Stokes image using a median filter method (described in Section 3.1.1 of Vanderwoude et al. 2024), achieving a sensitivity of 27 Jy beam in the MFS map. Subsequently, we performed source-finding in the resulting Stokes image using the pyBDSF (version 1.13.0) package (Mohan & Rafferty 2015), and identified 31,164 sources based on pixel brightness.
The Stokes cubes are convolved to a common beam of 21″ ( pc) over all frequency channels and then processed through the POSSUM pipeline (Van Eck et al., in preparation). The pipeline extracted Stokes , , and values across the entire frequency range for each of the 31,164 sources identified above, with the diffuse polarised emission subtracted from each source using an annulus median filter (Oberhelman et al. 2024). This step guarantees that our measured emission from the background sources is not contaminated by foreground diffuse emission, which can induce Faraday complexity3 and/or depolarisation4 (Ranchod et al. 2024). The median full-band RMS after diffuse emission subtraction, across all 31,164 sources, is 27.5, 19.8, and 19.6 Jy/beam in Stokes , and , respectively. The pipeline then applies RM synthesis (Brentjens & de Bruyn 2005; Heald et al. 2009) to every source using the RM-TOOLS (version 1.3.0) software (Van Eck et al. [129]). For the aforementioned band and channel width, we achieved a maximum detectable scale in Faraday space of rad m, a maximum Faraday depth rad m, and resolution (FWHM) of rad m (Brentjens & de Bruyn 2005; Dickey et al. 2019). We sampled the Faraday dispersion function over rad m with ten samples across the Rotation Measure Transfer Function (RMTF) FWHM, using inverse-variance channel weighting. The Stokes spectrum was modelled as a power law following the convention , with the polynomial order (zero, one, or two) selected per source by the Akaike Information Criterion (AIC). The resulting spectral indices for the full catalogue are shown in Figure 1. Within the catalogue, 16,783 sources had a reported spectral index of identically zero, which indicates that the POSSUM pipeline’s first-order fit to the Stokes spectrum did not significantly improve the AIC compared to the zeroth-order fit, and were excluded from our analysis.
Figure 1. Spectral index () versus Stokes signal-to-noise ratio (SNR) fit by the POSSUM pipeline (Van Eck et al., in preparation). The majority of sources beyond the Stokes SNR of 30 (blue, dashed vertical line) fall within the expected spectral index range of non-thermal sources. Median spectral index of the full, polarised and non-polarised distributions are shown as green, red and yellow lines respectively (see § 4.1 for further details). The uncertainties were determined via bootstrap resampling (1000 iterations), shown as the shaded region around the lines. Sources showing identically zero spectral index (totalling 16,783) are excluded from our analysis.
Neutral and Ionised Gas Density Tracers
To obtain magnetic field information from , details about thermal electron density are required. Usually, this is provided via H and/or neutral hydrogen (H I) data. Here, we use H I data to define the extent of the SMC and H to extract thermal electron density information.
Neutral hydrogen column density measurements, (cm), are taken from the ASKAP H I observations presented by Pingel et al. (2022) with a spectral resolution of km s and angular resolution of . The spatial coverage of these data is shown in Figure 2. Each background source is assigned an value using a nearest-pixel sampling approach in the H I data cube. For sources lying outside the ASKAP H I footprint or with spurious values at the beam edge, values are instead taken from the GASS H I survey (McClure-Griffiths et al. 2009).
Figure 2. RM grid across the SMC from the POSSUM Pilot II observations processed through the POSSUM pipeline, totalling 951 RMs. The area of the markers are linearly scaled by the absolute magnitude of the peak Faraday depth (), the largest at 419 rad m, coloured by the mean direction of the LOS component of the magnetic field, with red indicating towards us (positive RM) and blue away from us (negative RM). Markers of the RM sources are separated by Stokes -fitting model type, whilst left leaning tri markers represent polarised sources with no model fit, discussed in § 5. Upwards leaning tri markers represent nine sources designated as Faraday simple by -fitting but which show anomalous spectral structure (see Section A3). The RMs in this image and work are not corrected for the Milky Way RM foreground. The background greyscale map shows the H I column density of the SMC from the Pilot Survey of the Galactic ASKAP H I (GASKAP-H I) survey (Pingel et al. 2022).
We derive the Emission Measure (EM) from H observations. We use two sets of H observations, each at a different resolution. The first dataset is sourced from the Wisconsin H-Alpha Mapper (WHAM) survey, which observes diffuse H emission in the SMC at an angular resolution of 3600 (Smart et al. 2019) and covers the entire POSSUM field. The second is from the Magellanic Cloud Emission-Line Survey (MCELS), at a higher angular resolution of 5 and covering a central region of approximately , capturing more small-scale ionised structures compared to the WHAM survey (Winkler et al. 2015; Smith & Tean 1999). Both images are corrected for internal extinction by the SMC and the MW following Barger et al. (2013) and Smart et al. (2019). For polarised sources within the MCELS field of view, we assign associated emission from the MCELS data unless the data are determined to be non-physical (negative). In such cases, WHAM measurements are utilised instead. This captures a large range of extinction-corrected H intensity () observed across the SMC. The Emission Measure (EM) is then calculated as:
where is the electron temperature, assumed to be K (Smart et al. 2019) and is the path length of the ionised medium (see Equation (4) of Barger et al. 2013, for further details).
Methods
To characterise the and Faraday complexity of each polarised source, we employed two discrete techniques applied to the observed Stokes and polarisation spectra. First, we performed RM synthesis as implemented in the POSSUM pipeline, together with appropriate metrics to examine Faraday complexity from the RMCLEAN-reconstructed Faraday spectrum (Heald et al. 2009). Second, we conducted a parametric fit of theoretical models to the observed polarimetric spectra using QU-fitting (O’Sullivan et al. (2012); Sun et al. (2015); Anderson et al. (2016); Pasetto et al. (2018); Miyashita (2019); Thomson et al. [125]). Further analysis of depolarisation was explored in a statistical spatial framework.
RM Synthesis: Faraday Complexity Metrics
Currently, two primary metrics exist to classify a source as Faraday complex from its Faraday spectrum, . First, the second moment () of the RMCLEANed Faraday spectra provides an indication of how wide the spectrum is ([126] (Brown 2011; Anderson et al. 2015; Purcell & West 2017; Dickey et al. 2019; Livingston et al. 2021)). We normalise by the FWHM of the RMTF to account for the resolving power in Faraday depth space and use this new value () throughout the paper ([126] (Farnsworth et al. 2011; Kumazaki et al. 2014; Sun et al. 2015)). We adopt a threshold of to classify a source as Faraday complex (Vanderwoude et al. 2024).
The second metric of Faraday complexity is , developed by Purcell & West (2017) in pursuit of a more robust and discriminative measure of complexity. It is determined by passing the observed Stokes and spectra through RM synthesis to generate a Faraday spectrum, extracting , and the fractional polarisation . These quantities are passed into a Faraday simple (thin) model generating synthetic Stokes and . The synthetic spectrum is subtracted from the observed and structure beyond Gaussian noise in the residuals between them indicates a source is Faraday complex. A threshold of is used here to identify Faraday complex structure in the residuals. This complexity metric is a POSSUM pipeline product, and is detailed by Purcell & West (2017), Vanderwoude et al. (2024) and Van Eck et al. [129].
Stokes -Fitting
Stokes -fitting is a model-fitting technique that estimates the magneto-ionic structure along the line of sight by comparing observed Stokes and with modelled counterparts. To understand this approach, consider how the plane of polarisation interacts with various astrophysical environments. At one wavelength, the ellipticity of the light is characterised by the Stokes parameters , , , and . The Stokes and parameters define a polarisation vector in the plane, the magnitude of which gives the linear polarised intensity (PI). Stokes -fitting models how the plane of polarisation changes across the band. For Faraday complex sources experiencing depolarisation, the polarisation vector’s magnitude diminishes towards lower frequencies.
-Fitting Models
Models chosen for fitting were selected based on the most probable types of astrophysical environments resolvable with these data. We expect some manifestation of Faraday complexity either local or intrinsic to the source (Anderson et al. 2015; O’Sullivan et al. 2017). The models are labelled by their RM-tools nomenclature for consistency, and are described below. An isotropic homogeneous magnetised plasma in the foreground inducing Faraday rotation is classified as Faraday thin and is represented by model 1 (m1). When turbulence is present in the foreground magnetised plasma, the source is described by model 2 (m2). Model 3 (m3) describes two distinct polarised emission components subjected to the same level of turbulence in the foreground plasma. Model 4 (m4) describes a scenario where two emission components are subjected to different levels of turbulence in the foreground. Model 11 (m11) describes two separate regions of homogeneous magnetised plasma inducing two distinct modes of Faraday rotation, whilst Model 111 (m111) depicts three regions. The functional form of these models, their Faraday complexity classification and the number of free parameters (), are given in Table 1.
| Model | Description | Function () | |
| m1 | Faraday thin | 3 | |
| m2 | External Faraday Dispersion | 4 | |
| m3 | Two components with common external Faraday dispersion | 7 | |
| m4 | Two components with separate external Faraday dispersion | 8 | |
| m11 | Two Faraday thin | 6 | |
| m111 | Three Faraday thin | 9 |
Table 1. Table 1. -fitting models; model number corresponding to RM-TOOLS (Van Eck et al. 2026) designation, model description, functional form and number of free parameters (), where is the intrinsic degree of fractional polarisation, is the intrinsic polarisation angle, and is the Faraday depth component.
Best-Fit Model Selection
The parameter space of these models was searched with the nested sampling technique (Skilling 2004; Albert 2020), with parameters determined through the minimisation of , considering the observed polarised spectrum and the model polarised spectrum for model parameters (Miyashita 2019). Uniform priors were adopted for all parameters: , , , and . For multi-component models, additional constraints were imposed: , and between each pair of Faraday depth components.
To exclude any non-physical or “-fitting biased” results, the model fits of m1, m2 and m11 were scrutinised and filtered in accordance with the accurate model parameter spaces identified for the POSSUM band 1 data by Oberhelman et al. (2026), while models m3, m4, m111 were filtered in accordance with Ma et al. [70]. These filters ensure all valid solutions satisfy the following criteria: any component has an PI SNR , all Faraday depth components in multiple component models are separated by , and any component in Faraday dispersion models has an SNR in . From these valid solutions, we selected the best model based on Bayesian Evidence (), choosing the model with the largest value (Oberhelman et al. 2026).
Statistical Depolarisation
We examined the spatial distribution of polarised sources statically by subdividing the field into square degree bins, corresponding to , with CASA.5 Within these bins, we calculate the fraction of polarised sources per bin as:
where represents the number of polarised sources per bin and the total sources (both polarised and non-polarised) per bin. Assuming the source density adheres to a Poisson distribution, the error associated with each bin is calculated as , where denotes the number of sources, whether polarised or total. This error is then propagated to estimate the uncertainty in as:
This error assumes no correlation between the source densities and is thus the upper limit of the error. Only bins containing more than five polarised sources are considered to yield a reliable estimate of the uncertainty. We therefore exclude bins with polarised sources from our analysis (Figure 7 and 8). Additionally, we calculate the dispersion of peak Faraday depth (RM) per bin as,
Figure 7. Fraction of polarised sources, , against RM dispersion, , for the 2.25 deg bins containing polarised sources. Each bin is labelled by its column letter and row number, following Figure 6. Error bars show the propagated Poisson uncertainty on (§ 5.1) and the 16th–84th percentile bootstrap uncertainty on . Significance is assessed with one-tailed tests against a null distribution obtained by randomly redistributing all source positions across the field (10,000 iterations). Bins with significantly low (blue circles), significantly high (red squares), or both (purple star) have p-value < 0.05; open circles are not significant. The number of bins for each distribution is shown in parentheses in the legend. Grey shaded regions enclose 68 and 95 per cent of the null realisations, pooled over the plotted bins. Dashed lines mark the medians of the observed bins, and of Figure 6. The top and right histograms show the fraction of observed bins (black outline) and of null realisations (grey) in each histogram bin.
This is the standard deviation of peak Faraday depth per bin, where is the mean peak Faraday depth per bin. We repeat these calculations with bin sizes of and square degrees and find similar results; these are discussed in Appendix § 9.
Subscripts on , and denote either the bin area in deg (e.g. ) or a specific bin as labelled in Figure 6 (e.g. ). We do not correct the observed Faraday depths for the Galactic foreground; we instead assume the Galactic RM is approximately constant on the scales of our analysis bins. From the STAPS Milky Way RM reconstruction (Sun et al. 2025; Raycheva et al. 2025), the residual foreground variation across our bins is rad m, which is well below the statistical uncertainty on our on-SMC RM dispersion ( rad m). To demonstrate the statistical significance of the observed spatial structure, we performed a Monte Carlo spatial randomisation test. We uniformly redistributed all source positions across the field 10,000 times, preserving each source’s Faraday depth but randomising their locations; the total source count remains unchanged. For each iteration, we recalculated and in each bin.

Figure 6. Binned RM grid showing statistical depolarisation and RM dispersion. The upper panel shows the fraction of polarised sources (, Equation (4)) mapped per 2.25 deg across the SMC, with masked regions (indicating no polarised sources) shown in white. Each bin is labelled with a column letter and row number; the number of polarised sources is indicated in parentheses. Black contours at column densities of and cm, derived from ASKAP observations by Pingel et al. 2022, and red EM contours at 20 and 200 pc cm from Winkler et al. 2015 are overlaid, capturing the bulk of the small-scale thermal electron density. Notably, bins C2, D2 and D3 exhibit significantly reduced values compared to the rest of the field. The lower panel displays per bin; here, we present the spatial structure of the RM dispersion. With the same contours, bin labelling and polarised source counts as the upper panel.
Results
Polarised and Non-Polarised Source Selection
To ensure that we worked with a reliable catalogue of radio sources, we excluded sources with a Stokes SNR below 30 (see Figure 1), spectral index errors exceeding 0.5, spectral indices below , or fewer than 200 frequency channels.
From this parent sample, we constructed two sub-catalogues, polarised and non-polarised. We classified sources as polarised if they had a PI SNR greater than 7 and a fractional polarisation () between 0.01 and 0.72, with the lower bound set by the residual instrumental polarisation and the upper bound by the maximum theoretical fractional polarisation achievable for synchrotron emission (Rybicki & Lightman 1979). These criteria yielded a total of 1076 polarised sources from the RM synthesis results. However, our -fitting analysis (see § 4.2) revealed that 116 of these sources did not pass a model fit satisfying the criteria for a successful -fitting fit. So, we also excluded these from further analysis. An additional nine anomalous sources excluded from analysis possessed spectral structure inadequately modelled by -fitting (see Section A3 for a further discussion on these).
To ensure that the non-polarised catalogue traced the background extragalactic synchrotron population and not non-polarised sources intrinsic to the SMC, we removed 18 compact H II regions identified by Wong et al. (2012) and three planetary nebulae identified by Cotton et al. (2024) within 21″. To further isolate background extragalactic sources, we applied additional cuts removing sources with inverted spectra (), typically associated with optically thick thermal emission from compact H II regions (Hoare et al. 2007; Condon & Ransom 2016). We additionally removed sources with substantially extended morphologies (peak-to-total flux ratio ), which captures optically thin thermal sources () that are characteristically extended (Medina et al. 2019).
Then we removed duplicate detections both within and between the two catalogues using a 21″ matching threshold, retaining the source with the highest signal-to-noise ratio in the polarised intensity for polarised sources, and in Stokes for non-polarised sources. Where a source appeared in both catalogues, it was kept only in the polarised catalogue.
This yielded a final polarised catalogue of 951 sources, displayed in Figure 2, separated by their -fitting model assignment. This equated to approximately 24 RMs per square degree, a factor of improvement from the NVSS (Taylor et al. 2009) and a factor of 5 from SPICE-RACS [127]. This RM density is comparable to that achieved on the POSSUM works (Anderson et al. 2021; Vanderwoude et al. 2024). Figure 3 shows a histogram of the peak Faraday depth () for these polarised sources, separated into off- and on-SMC regions using an H I column density threshold of cm.

Figure 3. Histogram of peak Faraday depths separated by H I column density. The full polarised catalogue is shown as the black line. Sources considered off-SMC—those probing H I column densities less than —are shown in orange, while sources considered on-SMC, probing higher H I column densities, are shown in blue. These source are not Milky Way foreground corrected. A Kolmogorov–Smirnov test comparing the on-SMC and off-SMC distributions yields a statistic of 0.30 with a p-value of . Distribution statistics are shown in Table 2.
The resulting 1919 non-polarised sources are shown in Figure 4, exhibiting a median spectral index of . Previous observations of this nature suggest that the majority of these sources are unresolved Active Galactic Nuclei (AGN) or a single-lobe component (Norris et al. 2021; Filipović et al. 2021; Cotton et al. 2024; Khattab et al. 2025). We characterised source size using the Stokes flux density distribution. Sources were classified as compact when , and as extended otherwise. Table 2 lists the compact fraction for each -fitting sub-population; overall, 75% of the polarised catalogue is compact with this definition.

Figure 4. Locations of 1919 non-polarised background sources, coloured by peak brightness. Contours show H I column density () at 2 and from GASKAP-H I (Pingel et al. 2022).
| Catalogue | Sources Identified | Mean (rad m) | Median (rad m) | STD (rad m) | Mean | Mean | Compact (%) |
| Polarised | 951 | 76 | |||||
| Off-SMC | 811 | 75 | |||||
| On-SMC | 140 | 80 | |||||
| m1 | 727 | 73 | |||||
| m2 | 87 | 79 | |||||
| m3 | 42 | 79 | |||||
| m4 | 1 | 22 | – | – | 0.36 | 1.71 | 100 |
| m11 | 92 | 91 | |||||
| m111 | 2 | 50 |
Table 2. Faraday depth statistics of the polarised catalogue, comparing sources on and off the SMC and the results from -fitting. Columns display the population name, number of identified sources, the mean, median, and standard deviation (STD) of peak Faraday depths, mean , and percentage of the population that is compact. The uncertainties were determined via bootstrap resampling (1000 iterations). fitting identified 224 sources as Faraday complex. Weaknesses of complexity metrics from RM synthesis, fitting, m2 and are discussed in Section A4. On-SMC sources are those with a foreground H I column density exceeding cm.
We assessed the potential for per-channel (intra-channel) depolarisation arising from polarisation angle rotation across the finite width of a single frequency channel ( MHz). This effect is largest for the highest-RM sources at the lowest observed frequency, where the change in across a channel is greatest. Evaluating this worst case at the band edge (800 MHz), the rotation across a single channel is , and the largest value across the polarised catalogue ( rad m) is rad (8.6°). Per-channel depolarisation is therefore negligible for all sources (Gardner & Whiteoak 1966).
QU-Fitting Results
We applied the six different -fitting models in Table 1 to the 1076 polarised sources identified through RM synthesis. All sources were successfully fitted with a model. However, 116 sources failed to meet our reliability thresholds and were excluded from the catalogue and subsequent analysis (see § 3.3). We classified the remaining sources as either Faraday simple (727 sources), Faraday complex (224 sources) or anomalous (9 sources); we discuss anomalous sources in Section A3. Approximately 24% of the polarised catalogue exhibits Faraday complexity, with 14% of the polarised catalogue exhibiting Faraday dispersion (). The distribution of the model fits and statistics on the populations is detailed in Table 2 and shown across the SMC as a spherical Voronoi tessellation in Figure 5 and an RM grid-type structure in Figure 2. We found no statistical correlation between Faraday complexity and H I column density or EM, which we discuss further in Appendix A4.
Figure 5. Spherical Voronoi tessellation of polarised sources across the SMC, colour separated by their best fit -fitting model. Tessellation was constructed with scipy.spatial.SphericalVoronoi, with cell boundaries traced as great-circle arcs on a unit celestial sphere. To ensure finite and well-behaved cells near the edge of the field, artificial boundary points are placed just outside the analysis region and excluded from visualisation. Contours show H I column density () at 2 and from GASKAP-H I (Pingel et al. 2022). The tessellation highlights the spatial structure of Faraday complexity in the field. Each cell represents only the distance to the nearest source, not a region of influence; a length scale for should therefore not be inferred from this figure.
Spatial RM Dispersion and Statistical Depolarisation
Figure 6 displays per 2.25 deg bins across the SMC. Each bin is labelled with a column letter and a row number, and the number of polarised sources within each bin is indicated in parentheses. Contours of H I and EM for the SMC, representing the bulk of the neutral and ionised gas, respectively, are overlaid. The error () remains minimal and uniform across the centre of the field but increases near the field edges where the source density diminishes, with a median value of . Figure 6 (upper panel) highlights a notably lower value in bins encompassing the main H I body of the SMC, particularly in bins C2 and D2, compared to the surrounding regions, with a median value of . This spatial trend is maintained in both 0.25 and 1 square degree binning scales as shown in Figure A1, demonstrating that regardless of the chosen bin size, the same trend of a statistically lower polarised source density across the main H I body of the SMC holds. Figure 6 (lower panel) displays per 2.25 deg bins across the SMC, showing that bins with low coincide with regions of high RM dispersion.
Figure A1. Fraction of polarised sources, (top row), and peak Faraday depth dispersion, (bottom row), for 0.25 deg (left) and 1.00 deg (right) bins across the SMC. Black contours at column densities of and cm, derived from ASKAP observations by Pingel et al. 2022, and red EM contours at 30 and 200 pc cm from Winkler et al. 2015 are overlaid, capturing the bulk of the small-scale thermal electron density. Both quantities show the same spatial trends as the 2.25 deg maps in Figure 6, demonstrating that the trends persist across binning scales.
Figure 7 shows the relationship between and the peak Faraday depth dispersion for spatial bins containing more than five polarised sources, with statistical significance assessed through one-tailed permutation-based spatial randomisation tests (10,000 iterations). Scatter points show the 2.25 deg bins with their letter-number labels. Deviations are given in units of , the standard deviation of each bin’s (non-Gaussian) null distribution, alongside empirical one-tailed p-values. Bin D2 (purple star) deviates significantly from the null in both parameters (-value ): ( above the null mean) and ( below the null mean). Bins C2 and D3 (blue circles) exhibit a significantly low fraction of polarised sources, with ( below null, -value ) and ( below null, -value ), respectively. Bin A1 (red square) shows significantly elevated RM dispersion, with ( above null, -value ). The median fraction of polarised sources ranges over 0.343–0.391 across the three binning scales, and the median RM dispersion over 13.1–18.8 rad m for bins with . The median Faraday dispersion of the null distribution is 31.5 rad m, with most 2.25 deg bins falling below this value (Figure 7), because the Faraday dispersion is spatially concentrated across the main body of the SMC, randomising source positions distributes this structure across the field, resulting in a higher global null distribution.
Figure 8 shows the correlations between and with the median EM and column densities of polarised sources within each bin (for bins with ). decreases with increasing ionised and neutral gas column density, while increases with both tracers, with a stronger dependence on EM, consistent with expectations (Gaensler et al. 2005; Shah et al. 2025). For , we fit a linear-in-log model, finding and at the 2.25 square degree binning scale, with consistent slopes across all binning scales. The linear Pearson correlation coefficients are moderate, ranging from to for and from to for EM. For , we fit a power-law model, yielding and at the 2.25 square degree scale, again with consistent power-law indices across binning scales. The correlations for are systematically stronger than those for , with Pearson coefficients ranging from to for and to for EM. We find no discernible correlation between either or and the first or second moments of the H I emission spectra at any of the three binning scales.
Figure 8. Fraction of polarised sources, (top row), and Faraday dispersion, (bottom row), against the median H I column density, (panels (a) and (c)), and emission measure, EM (panels (b) and (d)), tracing the neutral and ionised gas, respectively. Only bins with are shown. The 2.25 deg bins are shown as open circles with error bars, with bins significant in the spatial randomisation test marked as in Figure 7; the 0.25 and 1.00 deg bins are shown as grey circles and green squares respectively. Lines show fits for each bin size: a linear-in-log model, , and a power law, , fitted in log–log space. Fit uncertainties are from bootstrap resampling (1000 iterations), with the shaded region showing the 16th–84th percentile interval of the 2.25 deg fit, whose slope and linear Pearson correlation coefficient, , are given in each panel. Panels (e) and (f) show the fitted and against bin area for both tracers; the horizontal line marks zero.
Observed statistical depolarisation trends are not dependent on the equal-area square grid adopted here, the same trend is observed in equal-area annuli. Additionally, the fraction of polarised sources is lower within the on-SMC region, with , compared to (see Figure 3); further on- and off-SMC statistics are given in Table 2.
Discussion
We now examine the occurrence of depolarisation across the SMC and the potential magneto-ionic structure responsible for it. We consider the complete statistical depolarisation of background sources, then attempt to decouple the magnetic field from the thermal electron density in an effort to identify which fluctuations in the Faraday-rotating medium are driving this depolarisation. Observations of Faraday complexity and its spatial structure are discussed in Appendix A4.
Statistical Depolarisation in the SMC
The identification of depolarisation through spatial statistics, enabled by the high density of this RM grid, provides a unique means to characterise both the amplitude and spatial scales of magneto-ionic turbulence in the SMC.
If the SMC’s RM field is isotropic and follows power-law scaling, then the upper and lower panels of Figure 6 likely sample different length scales of a single turbulent cascade. Under this framework, traces evidence of complete wavelength-dependent beam depolarisation from sub-beam-scale fluctuations ( pc) that produce Faraday dispersion , while traces larger-scale RM variations.
Elevated across the SMC shows a strong correlation with EM and at all bin sizes (Pearson to 0.8; Figure 8), indicating large-scale RM variations driven by the turbulent magneto-ionic medium, well beyond the physical scales traced by sub-beam Faraday dispersion. The fraction of polarised sources, , is also anticorrelated with both tracers, but more weakly ( to ). Higher SMC gas densities therefore give rise to both enhanced RM dispersion and fewer polarised sources, showing that the SMC is depolarising background sources. The reduced polarised source densities in bins D2, D3, and C2 indicate strong beam depolarisation, demonstrating that RM fluctuations exist on both large and small scales. The contrast in correlation strength arises naturally if the electron density is relatively uniform along the LOS on the scale of magnetic fluctuations and acts primarily as a scaling for RM fluctuations (Section 5.2): it sets the amplitude of RM fluctuations on all scales, so the RM dispersion between sightlines closely tracks EM, whereas the sub-beam Faraday dispersion that depolarises sources is much weaker and follows the gas density only loosely, as expected if it is set by the magnetic fluctuations mainly.
Bin Features
Source finding plays an important role in estimating , particularly regarding local noise levels and potential continuum contamination from SMC emission. Although we performed source finding on a diffuse-subtracted Stokes image to minimise extended emission, compact sources and non-thermal emitters may still be detected, leading to source count overestimation along the Bar and Wing region, particularly for non-polarised sources. To mitigate this, we applied spectral index cuts (see § 4 and Figure 1) and removed known compact H II regions and planetary nebulae. This established an approximate background uniformity in the non-thermal non-polarised distribution, finding mean counts of 12.3 ± 4.5 (0.25 deg bins, ), 45.6 ± 14.8 (1.0 deg bins, ), and 122.5 ± 18.4 (2.25 deg interior bins, , excluding low-count edge regions). This uniform distribution of non-polarised sources indicates a successful isolation of extragalactic sources, validating the map presented in Figure 6.
Bin F3, containing five polarised sources, initially appears depolarised but exhibits low . Its value does not persist across spatial scales, suggesting scale-dependent behaviour likely due to beam-edge low-number statistics rather than physical depolarisation. Deeper observations with higher source densities would enable smaller bin sizes without compromising statistical reliability. Furthermore, the complete coverage of the Magellanic clouds by the POSSUM survey would mitigate any edge effects (Gaensler et al. 2025).
The elevated Faraday dispersion in bin A1 is driven by a single Faraday simple source, J014316.6−713644 with rad m, which lies ( kpc, or ) from the centre of the background galaxy cluster 1eRASS J014318.8−713652 (photometric , ; Bulbul et al. 2024), also detected by Planck as PSZ2 G297.16−44.90 (Planck Collaboration et al. [97]). Its observed Faraday rotation is therefore likely dominated by the intra-cluster medium; excluding it reduces to 13.7 rad m, consistent with the field, and removes its apparent outlier behaviour in the correlations with the neutral and ionised gas tracers (Figure 8).
What is driving depolarisation?
The observed spatial statistical depolarisation arises from a turbulent magneto-ionic structure within the beam. This turbulence is characterised by the strength () and coherence scale () of the random magnetic field component, along with the number density fluctuations () and coherence scale () of the thermal electrons, each contributing to the observed Faraday depth. The filling fraction of each component, the fraction of the LOS occupied by coherent structures, may differ between the two. The relative magnitudes of these properties influence which component dominates the depolarisation process (Xu & Zhang 2016). For example, in the Milky Way, the fluctuations have a much larger scale ( pc) compared to the magnetic field fluctuation scale ( pc) (Armstrong et al. 1995; Dhakal & Seta 2025). To determine the primary driver of the depolarisation, these components must be evaluated separately6.
In the SMC, the prior observations favour the magnetic field as the primary driver. Although working with a spatially sparser dataset (Livingston et al. 2022), Seta et al. (2023) found approximately constant across the pc correlation length of the random field, suggesting that on these scales the RM fluctuations are primarily set by the magnetic field rather than by the thermal electron density. This suggests a picture in which the thermal electron fluctuations are smooth along the LOS, concentrated in one or two large ionised regions (large ), while the magnetic field is tangled on much smaller scales, with frequent changes in strength and orientation (small ). In such a medium, cannot vary consistently with the small-scale field, and we therefore attribute the Faraday dispersion to the random magnetic field throughout.
The relationship between fractional polarisation and EM can be understood through the framework of Gaensler et al. (2005), who studied the LMC. Depolarisation arises not from the mean Faraday rotation along a LOS, but from the dispersion in rotation measure across the beam. In the Gaensler et al. (2005) formulation, it is assumed that is uncorrelated with the magnetic field, that the filling fraction of thermal electrons () is small, and that the random component of the magnetic field is constant across the LMC, such that , where . This follows from a random walk argument: a sightline of length crosses independent cells of size , each adding a Faraday depth of with random sign, where , so that with (the number of ionised cells and the EM both scale with , so is unchanged). Here, EM enters not as the driver of depolarisation but as a scaling factor: regions with higher electron density convert a given level of magnetic field fluctuations into larger RM fluctuations, and hence stronger depolarisation. In Figure 9, we show the fractional polarisation of our polarised catalogue as a function of EM. Sources are binned logarithmically with four bins per order of magnitude, and we present the median and mean values for each bin.
Figure 9. Peak fractional polarisation against emission measure. The polarised catalogue coloured by their PI SNR (circles), with error bars. Sources are binned four per order of magnitude in EM, bin means (white circles) and means (black triangles) are shown with 68% intervals. Both lines are maximum-likelihood fits to the individual sources in , the blue dashed line is a power law, . The red solid line is the Burn model, , i.e. , with m the band average. Shaded regions are the 68% bootstrap ranges of each fit. The black dashed line marks .
We fit this model to the unbinned data using Burn’s law for external Faraday dispersion (beam depolarisation) (Burn 1966; Tribble 1991), , where is the POSSUM band average. With , this becomes , so should fall linearly with EM at a rate set by . These measures show much Faraday dispersion the SMC produces at a given EM. We find and (Figure 9), with uncertainties determined via bootstrap resampling (1000 iterations). The implied Faraday dispersion is modest: at , rising to at the highest EMs in our sample (), where it halves . This is no larger than the spread between individual sources, which differ by a factor of at any EM ( in ), so the SMC’s imprint is subtle.
-fitting offers an independent route to , measuring for individual sources rather than inferring it from the population trend. We use the 102 compact sources best fitted by models m2 and m3 (69 and 33, respectively), which include external Faraday dispersion (Table 1; the single m4 source is excluded). Assuming that for the 88 sources off the SMC arises from intrinsic and Milky Way dispersion, we use them to set a floor of . The 14 sources behind the SMC have a higher median, . Independent Faraday screens add in quadrature, , and fitting the individual sources with yields , consistent with the Burn’s law fit (Figure 9). Subtracting the floor in quadrature from the median gives an SMC contribution of along these sightlines, whose median EM is , equivalent to .
Our measured value of is a factor of lower than that found in the LMC (; Gaensler et al. (2005)), indicating substantially weaker Faraday dispersion in the SMC at a given EM. For , the LMC value of implies , sufficient to completely depolarise background sources across our observing band, compared with for our measured . Given that , this difference is consistent with a weaker random magnetic field, a smaller turbulent cell size, or both, in the SMC. However, Gaensler et al. (2005) measured at 1.4 GHz, where Faraday dispersions of up to remain detectable, using sightlines largely through the LMC itself, whereas at our lower frequency sources with are depolarised below the detection limit, and our sample extends well beyond the main body of the SMC (Figure 2). The spatially coherent statistical depolarisation observed across the SMC (Figure 6) indicates that the magneto-ionic turbulence is not uniform across the SMC; sightlines through the most turbulent gas are not observed (Figure 8), biasing our estimate of towards less turbulent environments. Our measured therefore characterises the less turbulent gas through which polarised background sources survive, and is effectively a lower limit.
Translating into a field strength requires knowledge of the cell size. For , , normalised to the 0.5 pc cells that Gaensler et al. (2005) adopted for the LMC: larger cells give fewer but larger steps in the random walk, so the same dispersion requires a weaker field. Depolarisation requires many cells within each pc beam, so pc and . In turbulence with a continuous range of scales, this bound becomes an estimate: the Faraday dispersion within a beam is dominated by the largest eddies that still fit inside it, so the effective cell size is close to the beam itself and on pc scales. This estimate assumes that H traces the Faraday-rotating gas, and becomes an upper limit if fluctuations in electron density also contribute to .
Seta et al. (2023) also measured the spectrum of the turbulent field on larger scales, using RM structure functions. They find a correlation length of pc, a random-field strength of , POS RM fluctuations between independent sightlines reaching , and a magnetic power spectrum of slope below . Because the SMC is far deeper along the LOS ( kpc; Smart et al. (2019)) than , every sightline crosses at least independent cells, so RM differences accumulate as a random walk; with nearly constant, the RM structure function then rises as below , with the same slope as the magnetic power spectrum. Their sightlines have a mean H intensity of R, corresponding to EM pc cm for K (Reynolds 1977). If the field keeps this spectrum down to smaller scales, the RM dispersion within a beam of size , which goes as the square root of the structure function, is . For our 6 pc beam, this predicts rad m, whereas at EM pc cm our Burn’s law fit gives rad m, four times less. The same holds for the field strength: their G assumes a filling factor of 0.4, equivalent to G at 160 pc in our formulation, in which the filling factor cancels; with , this falls only to G at 6 pc, against the G we infer. Reconciling the two requires a steeper magnetic power spectrum between the beam and 160 pc, with a slope of , steeper than both their fitted slope and Kolmogorov (), but within of their measurement. Our depolarisation therefore probes the small-scale end of the turbulence they measured, and indicates less Faraday-rotating power on parsec scales than a single power law from 160 pc implies.
However, these interpretations are sensitive to the three-dimensional structure of the SMC, which remains uncertain (Mathewson et al. 1987; Ripepi et al. 2017; Murray et al. 2024). The SMC likely comprises two components separated by kpc, implying that the characteristic length scales driving depolarisation may vary spatially. A more complete understanding of the SMC’s geometry is needed to robustly disentangle the respective contributions of and , a goal of future work.
Nevertheless, the overall picture of magnetic-field-driven depolarisation is consistent with other extragalactic studies, which through modelling and multi-wavelength analysis typically attribute depolarisation to a turbulent magnetic field (Fletcher et al. 2011; Basu et al. 2017; Kierdorf et al. 2020). This is further supported by Galactic observations indicating that the thermal electron distribution is more uniform across scales, while the magnetic field structure is more influential (Marsch 1991; Minter & Spangler 1996; Haverkorn et al. 2008; Dhakal & Seta 2025). High angular resolution observations of , such as those anticipated from the Sloan Digital Sky Survey V Local Volume Mapper (Drory et al. 2024), can provide the spatial detail needed to constrain the role of thermal electron fluctuations in driving depolarisation.
Conclusion
Using the Polarisation Sky Survey of the Universe’s Magnetism (POSSUM) Pilot II survey on the Australian SKA Pathfinder (ASKAP), we present a high-density rotation measure (RM) grid of 951 broadband polarised background radio sources ( RMs per square degree) to probe the magneto-ionic structure of the Small Magellanic Cloud (SMC). Utilising the radio polarimetric analysis algorithms of RM-synthesis and Stokes -fitting, with each source fit using six distinct Faraday models, this work presents, to our knowledge, the first large-scale application of -fitting to such a dense RM grid across this broad a bandwidth.
We characterised the SMC’s magneto-ionic properties using two spatial measures - the fraction of polarised sources - and RM dispersion between sources - complemented by RM dispersion towards source where observed. We identified spatially coherent depolarisation () signatures across the SMC that correlated with enhanced plane-of-sky RM fluctuations () (Figure 6), revealed through a statistical comparison of polarised and non-polarised non-thermal source densities on a per-square-degree basis. The absence of polarised sources across the main body of the SMC is statistically significant, with depolarisation strengthening in regions of elevated RM dispersion. Both quantities are correlated with emission measure and neutral hydrogen column density, and persistent across a range of binning scales. These trends demonstrate that the magneto-ionic medium of the SMC drives both enhanced RM dispersion and depolarisation of background sources. Population-level modelling of beam depolarisation with emission measure indicates that the SMC produces times less Faraday dispersion than the Large Magellanic Cloud at the same emission measure, although differences in observing frequency and sightline distribution likely contribute to the discrepancy. Additionally, complete depolarisation of sources behind the most turbulent gas of the SMC biases our fit towards less turbulent sightlines.
Faraday complexity was detected in of the RM catalogue; however, no coherent spatial structure was observed in the complexity distribution. We attribute this complexity primarily to internal source structure or an extended halo surrounding the SMC, rather than to the magneto-ionic medium of the SMC main body. Standard RM synthesis-based complexity metrics routinely used in the literature ( and ) failed to recover the full population of complex sources identified through -fitting, with their classification accuracy strongly dependent on polarised intensity signal-to-noise ratio (see Appendix A4).
Taken together, these results provide compelling evidence that the SMC depolarises background radio sources through a turbulent magneto-ionic medium on scales smaller than the beam ( pc). This depolarisation is most likely driven by small-scale fluctuations in the magnetic field, which on these scales must have a random field strength of G. This work demonstrates the capability of dense RM grids to resolve the detailed line-of-sight magnetic field structure of the interstellar medium.
Acknowledgements
J. M. P thanks Kathleen A. Barger, Sui Ann Mao, and Rainer Beck for their thoughts and comments on this work, and Luka Mijnarends and Eric Muller for their support. We thank the anonymous reviewer, whose comments have improved the quality and rigour of this paper. This research was partially funded by the Australian Government through an Australian Research Council Australian Laureate Fellowship (project number FL210100039 awarded to N. M. McG.) and a Discovery Project (DP220101558). J. M. P acknowledges the German Academic Exchange Service (DAAD). Y. K. M. is partially supported by the Federal Ministry of Education and Research (BMBF) in Germany via the German Academic Exchange Service (DAAD) Project-based Personnel Exchange Programme (PPP). A. S. is supported by the Australian Research Council through the Discovery Early Career Researcher Award (DECRA) Fellowship (project DE250100003) funded by the Australian Government and the Australia-Germany Joint Research Cooperation Scheme of Universities Australia (UA–DAAD, 2025–2026). S. L. J acknowledges support from the UK Research and Innovation (UKRI) Frontiers Research Grant [EP/X026639/1]. J. D. L is funded under the Mapping Magnetic Fields with INterferometry Down to Event hoRizon Scales (M2FINDERS) project. The M2FINDERS project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 101018682). S. P. O acknowledges support from the Comunidad de Madrid Atracción de Talento program via grant 2022-T1/TIC-23797, and grant PID2023-146372OBI00 funded by MICIU/AEI/10.13039/501100011033 and by ERDF, EU.
This scientific work uses data obtained from Inyarrimanha Ilgari Bundara, the CSIRO Murchison Radio-astronomy Observatory. We acknowledge the Wajarri Yamaji People as the Traditional Owners and native title holders of the Observatory site. CSIRO’s ASKAP radio telescope is part of the Australia Telescope National Facility (https://ror.org/05qajvd42). Operation of ASKAP is funded by the Australian Government with support from the National Collaborative Research Infrastructure Strategy. ASKAP uses the resources of the Pawsey Supercomputing Research Centre. Establishment of ASKAP, Inyarrimanha Ilgari Bundara, the CSIRO Murchison Radio-astronomy Observatory and the Pawsey Supercomputing Research Centre are initiatives of the Australian Government, with support from the Government of Western Australia and the Science and Industry Endowment Fund.
Data Availability
This paper makes use of raw data products that are publicly available through the CSIRO ASKAP Science Data Archive (CASDA; (Chapman et al. (2017); Huynh et al. (2020))). The processed POSSUM data used in this work is subject to the guidelines of the POSSUM survey and will be shared upon reasonable request to the corresponding author.
Supplementary Material
Statistical Depolarisation at smaller scales
Figure A1 present at smaller binning scales compared to those in Figure 6, at 0.25 (left) and 1.0 (right) square degrees respectively. These demonstrate that the spatial statistical depolarisation trend is preserved across binning scales, in addition to Figure 8.
RM Jitter
RM jitter is the additional scatter in measured rotation measures produced by unresolved Faraday rotation fluctuations in turbulent plasma local to the source on sub-beam scales, and is most evident statistically in populations of sources with low fractional polarisation ((Stil 2025)). Following the approach of Vanderwoude et al. (2024), who identified RM jitter potential signatures by comparing the interquartile range (IQR) of RM distributions between weakly polarised ( fractional polarisation) and strongly polarised () source populations, we find no significant difference between the two samples in our field: the IQR for low- and high-polarisation sources is 30.96 rad m and 29.66 rad m, respectively, in contrast to the difference reported by Vanderwoude et al. (2024). This similarity persists when restricting to off-SMC sources (28.01 vs 26.46 rad m). Instead, the dominant source of RM scatter is the large-scale Faraday structure of the SMC itself, as evidenced by the substantially higher IQR for on-SMC sources (54.57 rad m) compared to off-SMC sources (26.99 rad m). We conclude that RM jitter does not significantly contribute to the observed RM dispersion in this field, and does not impact the conclusion that the SMC is depolarising background sources.
Anomalous Sources
In this section we provide a brief discussion of the spectropolarimetrically anomalous sources within the catalogue. We define these as sources that were successfully fitted by Stokes -fitting using a Faraday-simple model (m1), but which possess a . A high indicates that the observed fractional Stokes and spectra deviate significantly from the behaviour expected for a single Faraday-thin component, suggesting Faraday complexity. There are nine such sources in the field, which we list in Table A1.
| Source Name | RA (deg) | Dec (deg) | (rad m) | (%) | (deg) | SNR | SNR | Compact | ||||
| J013243.9-734413 | 23.1829 | -73.7372 | 2.39 | 65.9 | 1509.0 | 0.40 | 1.46 | 1500.6 | Y | |||
| J010255.7-743536 | 15.7323 | -74.5936 | 4.23 | 54.6 | 824.6 | 0.19 | 1.12 | 1196.4 | Y | |||
| J004936.2-741547 | 12.4009 | -74.2631 | 14.05 | 242.6 | 1198.5 | 0.21 | 2.24 | 865.4 | Y | |||
| J002008.4-751225 | 5.0349 | -75.2072 | 4.24 | 27.0 | 375.9 | 0.33 | 1.19 | 791.0 | Y | |||
| J014235.7-723001 | 25.6487 | -72.5004 | 6.00 | 107.2 | 1576.0 | 0.16 | 1.09 | 1368.0 | Y | |||
| J005533.6-723126 | 13.8902 | -72.5241 | 1.63 | 25.1 | 709.3 | 0.51 | 1.27 | 1243.6 | Y | |||
| J005523.4-721056 | 13.8474 | -72.1823 | 1.15 | 54.1 | 1921.3 | 1.37 | 3.63 | 240.1 | Y | |||
| J004413.8-724302 | 11.0577 | -72.7173 | 2.74 | 75.8 | 1456.8 | 0.29 | 1.02 | 1654.6 | Y | |||
| J003401.1-702626 | 8.5045 | -70.4407 | 5.03 | 234.1 | 2014.3 | 0.27 | 1.62 | 1723.7 | Y |
Table A1. Anomalous sources best fit by QU-fitting with Model 1 as Faraday simple but with . These sources are not considered in the analysis of the main text. This table identities the nine such sources in this catalogue with their source Name, Right Ascension, Declination, peak Faraday depth from RM synthesis, fractional polarisation, de-rotated polarisation angle, signal-to-noise ratio in polarised intensity, signal-to-noise ratio in Stokes , normalised m2, (see § 3.1), log Bayesian evidence and source compactness (a source is consider compact if it speak flux is within 10% of its total flux from the source finder pyBDSF (compact: Y, extended: N) (see § 2).
Several physical and observational effects may give rise to such anomalous behaviour. At the angular resolution of these observations, the most likely cause is unresolved internal structure within the emitting component or Faraday-active plasma local to the source. The -fitting models applied in this work examines multi-component Faraday structure and external Faraday dispersion; however, it does not include models that produce repolarisation, the phenomenon in which fractional polarisation increases with wavelength, sometimes referred to as inverse depolarisation ((e.g. Homan et al. 2002; Hovatta et al. 2012; Homan 2012; Farnes et al. 2014)). A variety of physical scenarios have been proposed to produce repolarisation. For example, Homan (2012) proposed a model in which structural inhomogeneities in a jet magnetic field and internal Faraday rotation give rise to this effect. In this model, differences in the magnetic field structure between the near and far sides of the jet lead to partial cancellation of polarisation along the line of sight, resulting in low net polarisation at short wavelengths. At longer wavelengths, internal Faraday rotation, which grows with wavelength squared, acts to align the polarisation contributions from the two sides of the jet, reducing the cancellation and producing a net increase in fractional polarisation. One might also expect repolarisation to arise naturally in unresolved, optically thick core regions where different wavelengths probe different physical depths within the jet ((Konigl 1981)).
We show one such repolarising source, J005523.4−721056, as an example in Figure A2. The spectral structure of this source exhibits a clear repolarisation signature at longer wavelengths. Our -fitting procedure initially excluded this feature and fits only the longer-wavelength data, which correspond to the peak Faraday depth recovered by RM synthesis, yielding rad m. This source was also identified by Livingston et al. (2022), who reported RM rad m from 1.4–3.0 GHz (see their Figure 46 in supplementary material), consistent with our measurement. We note that the repolarisation characteristics of this source bear a superficial resemblance to those sometimes seen in pulsars; however, the MFS image reveals that the emission is associated with a bright component of a radio-loud AGN with a morphology consistent with an FR II galaxy.
Figure A2. Anomalous source J005523.4-721056, exhibiting repolarisation features. Top left: Stokes intensity from the MFS image. The synthesised beam is shown in the lower left corner, with a major axis. White contours indicate positive emission at levels of , , , , , and , while magenta dashed contours trace negative emission at and , where µJy/beam. Top right: Faraday spectrum showing the dirty (dashed black), clean (solid red), and residuals (solid blue). The vertical dot-dashed line marks the peak Faraday depth at rad m. Bottom left: Spectral properties as a function of wavelength: Stokes intensity (top), fractional Stokes and with fractional polarisation (middle), and polarisation angle (bottom). A secondary frequency axis is shown along the top. Bottom right: Fractional Stokes versus , colour-coded by wavelength.
Interpretation of Faraday complexity
Astrophysical Faraday Complexity
The observed Faraday complexity in this study does not appear to arise primarily from the SMC main body, but rather from effects intrinsic to the background sources, or possibly from some extended structure around the SMC. The spherical Voronoi tessellation in Figure 5, reveals no distinct spatial structure in the distribution of Faraday complex sources or of those exhibiting Faraday dispersion. We find no correlation between Faraday complexity and H I column density or H emission.
Previous studies (e.g., (Law et al. 2011; Rudnick & Owen 2014)) have demonstrated that multi polarisation component sources can arise from beam effects or limitations in source-finding algorithms used in polarimetric surveys. In particular, consider the case of Fanaroff-Riley Class II (FR II) sources, for which the polarised emission from the bipolar jets can individually experience dissimilar Faraday rotation from their local magneto-ionic medium ((Garrington et al. 1988; Laing 1988a; Law et al. 2011)). If such FR II sources are (at least partially) spatially resolved, our extraction method that identifies the peak brightness at the pixel level should be able to identify the two lobes as separate spatial components, rather than combining them into a single component. However, if the FR II source is unresolved, the two spatial components can be combined to produce a multi-component Faraday spectrum that is related to the physical structures within/surrounding the source. As shown in Table 2, 91% of m11 sources and 79% of m3 sources are considered compact, the single m4 sources, along with one m111 are also compact. The investigation of m111 and m4 sources is beyond the scope of this work but they are likely unrelated to the SMC. The 24% Faraday complex fraction of this sample is most likely a reflection of the observational parameters, as Faraday complexity is known to depend on frequency coverage and Stokes signal-to-noise ratio ((Law et al. 2011; O’Sullivan et al. 2017; Oeberlmann et al. 2026)).
The appearance of Faraday complexity in the sample is likely a result of source structure, reaffirmed in Figure A3, which examines the fractional difference in fractional polarised intensity and difference in Faraday depth () between the components of models m3, m4, m11 and m111. If the multi-component structure arose from SMC turbulence, we would expect component separations to follow predictions based on the SMC’s magneto-ionic properties. Seta et al. (2023) extrapolated the
Figure A3. Difference in the fractional polarisation of each component for multi-component models against their difference in Faraday depth. For two-component models (m3, m4, m11), and the fractional polarisation difference is calculated as . For the three-component model 111, represents the mean pairwise difference among the three components: , and the fractional polarisation difference is the mean of the three pairwise ratios minus 1: . Median lines of and are shown as dot-dashed black and red lines respectively. The expected global turbulence for the SMC from Seta et al. 2023 at approximately is displayed as the blue dashed line. O’Sullivan et al. 2012 identified the two-component AGN PKS B1610-771 (blue star), along with a three-component AGN PKS B1039-47 (orange cross). Reference lines of two and four times the difference in fractional polarisation between components are shown as the purple and brown dashed lines respectively. Sources lower in fractional differences have components similar in polarised sensitivity.
RM standard deviation at rad m with RM structure functions using the data of Livingston et al. (2022), at the break scale of pc. This prediction is shown as the blue dashed line in Figure A3 and represents the expected for a second component if induced by global SMC turbulence. However, our observed component separations do not follow this prediction, but rather follow an uncorrelated scatter, suggesting that the multi-component structure is independent of the SMC magneto-ionic medium. Within this field, we capture more RM measurements across the entire Magellanic ecosystem than ever before. However, there exists the possibility that separating sources on and off the SMC is not feasible if the Magellanic ecosystem contains ionised Faraday-rotating gas beyond the ionised halo of the SMC, extending towards the Stream, Bridge, and Bar of the ecosystem. If this gas were Faraday-rotating and detectable at these frequencies, the Faraday dispersion observed in Model 2 from -fitting could arise from an extended envelope of magnetized gas beyond the local boundaries of the SMC. Additionally, the extent of interaction with the Milky Way’s halo remains unclear (D’Onghia & Fox 2016; Choi et al. 2022; Mishra et al. 2024; Carr et al. 2025). Hence, with our SMC data alone, we cannot decisively distinguish the astrophysical origin of Faraday complexity.
More broadly, the physical origin of multiple Faraday components is not always clear. They may arise from intrinsic source structure, magneto-ionic material local to the source environment, or intervening foreground media (Bridle & Perley 1984; Perley et al. 1984; O’Sullivan et al. 2017). In the case of Figure A3, one might ideally expect distinct clustering or correlation between Faraday components in the versus space that could help disentangle their origin. For instance, the Laing-Garrington effect predicts systematic asymmetries in depolarisation between the approaching and receding lobes of double radio sources (Garrington et al. 1988; Laing 1988b), while sources of similar morphology, such as radio galaxies or AGN with unresolved bipolar jets aligned along the LOS might be expected to exhibit comparable properties in and fractional polarisation difference, modulated by inclination angle (O’Sullivan et al. 2017). In a sufficiently large and representative population sample drawn from full-sky observations, sources of distinct morphological and astrophysical types may therefore occupy well-defined regions within this parameter space. Additionally, the observed Faraday depth of a source is subject to cosmological dilution, meaning that intrinsic Faraday rotation at the source redshift is reduced in the observer frame, introducing further scatter for sources at different redshifts (Kronberg et al. 2008; Durrer & Neronov 2013. One might also expect the large-scale group or cluster environment to contribute additional Faraday complexity for sources embedded in dense magneto-ionic media, though we assume this is not a dominant effect for our sample. Beyond this, small-scale Faraday rotation fluctuations from the Milky Way on angular scales comparable to the data must also be considered. With the deeper fields and increased broadband frequency coverage that are becoming prevalent in the radio community, particularly with the advent of the SKA, filling in this parameter space becomes a realistic possibility.
Identification of Faraday Complexity
The motivation for Faraday complexity classification arises from the need to investigate the Faraday-rotating plasmas and in our quantification of the . In Figure A4, we present the two Faraday complexity metrics derived from RM synthesis, and , for all polarised sources; these sources are separated according to their best-fit model obtained from -fitting. Using complexity thresholds suitable for POSSUM data (see Section 4.3.4 in Vanderwoude et al. 2024), namely and , the lower-left quadrant of the figure corresponds to sources classified as Faraday simple by both metrics, while the upper-right quadrant corresponds to sources deemed Faraday complex by both metrics.

Figure A4. Normalised second moment () versus for all polarised sources, coloured by their PI SNR. Sources are distinguished by their best-fit -fitting models: circles for Model 1, diamonds for Model 2, squares for Model 3, triangles for Model 11, and crosses for Model 111. Magenta coloured tri markers indicate sources designated as model 1 with these sources removed from the catalogue are discussed in Section A4. Horizontal and vertical lines indicate the thresholds for Faraday complexity classification, set at and . 85% of sources are within the lower left quadrant.
When holding -fitting as the ground truth, several sources are misclassified by both metrics; however, the misclassification rate appears to be higher for than for . The PI SNR of the source is also crucial for correct classification amongst the RM synthesis tools. For sources with SNR PI , correctly classify 88%, compared to 38% for . To reach 90% accuracy, requires an SNR PI , while requires an SNR PI . However, at low PI SNR, -fitting is also biased toward simpler models, so the trend of increasing agreement at higher PI SNR cannot be attributed solely to improved RM synthesis metric performance.
While was an effective diagnostic of strongly Faraday-complex behaviour with moderate PI SNR, it was less sensitive to mild complexity. Consequently, sources exhibiting high that were classified as Model 1 (Faraday simple) by -fitting, indicating Faraday complexity that could not be captured by our multi-component models. We identified nine such sources in the field and excluded them from further analysis, as no suitable model adequately described their polarisation behaviour. These sources are marked as anomalous in Figure 2 and Figure A4. One example exhibiting repolarisation is presented in Appendix A3. The spectral characteristics of these anomalous sources likely arose from intrinsic structure within the emitting volume (Anderson et al. 2015), which our externally screened models could not reproduce.
The 116 sources that were detected in RM synthesis but failed the -fitting reliability criteria were too faint, with median SNR Stokes I and PI SNR approximately half those of successfully fitted sources. Despite being intrinsically polarised (median fractional polarisation of 6.3%), their low flux densities precluded reliable model fitting. The spatial distribution is random across the field (see Figure 2).
References
- [67]Li H.-B., et al., 2015, Nature, 520, 518
- [70]Ma Y. K., Mao S. A., Stil J., Basu A., West J., Heiles C., Hill A. S., Betti S. K., 2019, MNRAS, 487, 3432
- [71]Ma Y. K., et al., 2023, MNRAS, 521, 60
- [97]Planck Collaboration et al., 2016, A&A, 594, A27
- [125]Thomson A. J. M., et al., 2021, MNRAS, 507, 3495
- [126]Thomson A. J. M., et al., 2023, Publ. Astron. Soc. Australia, 40, e040
- [127]Thomson A. J. M., et al., 2026, arXiv e-prints, p. arXiv:2605.16917arxiv.org/abs/2605.16917
- [129]Van Eck C. L., et al., 2026, RM-Tools: Software for Analyzing Polarized Radio Spectra (arXiv:2601.20092), https://arxiv.org/abs/2601.20092








