CHANG-ES XL: Magnetic Field Structures in the Disk and Halo of NGC 891

This paper presents new Karl G. Jansky Very Large Array (VLA) S-band (2-4 GHz) observations of the edge-on spiral galaxy NGC 891, complemented by C-band data, to investigate the structure of its radio continuum halo. NGC 891 is a nearby edge-on spiral galaxy located at RA (J2000) = 02h 22m 32.91s, DEC (J2000) = +42° 20′ 53.95″, at a distance of 9.1 Mpc, with an inclination of 89.8° and a Hubble type of Sb. It has long been considered a close analog, or twin of the Milky Way owing to its similar optical luminosity, morphology, and rotation velocity, though it exhibits a somewhat higher star formation rate.

Two S-band observations were made on June 29, 2021 and August 13, 2021, each with 3.13 hours on source. The observations were carried out with two pointings in interleaved mode. 3C48 was used as the primary flux calibrator and polarization angle calibrator, with 3C84 as unpolarized calibrator, and J0251+4315 as the phase calibrator. Polarization calibration was performed in three main steps: determining the instrumental delay between the two polarization outputs, solving for the instrumental polarization, and solving for the polarization position angle using 3C48.

The S-band data were imaged with multi-scale tclean with the gridder set to mosaic. Three different angular resolutions were obtained using uv-tapering: high resolution (4.92″ × 4.65″ FWHM, rms noise σI = 3.4 µJy Beam−1), intermediate resolution (10.10″ × 9.83″ FWHM, rms noise σI = 5.4 µJy Beam−1), and low resolution (20.65″ × 18.83″ FWHM, rms noise σI = 12.5 µJy Beam−1). The high-resolution image clearly shows the thin disk of NGC 891 and allows identification of compact emission features along the midplane, while the lower-resolution maps are more sensitive to diffuse halo emission, extending several kpc above and below the disk plane.

The 3 GHz flux density of NGC 891 was measured to be 366 ± 1 mJy from the 20″ image corrected for primary beam attenuation. Comparing with interpolated flux densities from 1.5 GHz (737 ± 37 mJy) and 6 GHz (252 ± 27 mJy), the VLA observations recovered approximately 85% of the total flux density at 3 GHz.

For the RM synthesis analysis, the Stokes Q, U, and I cubes in S-band and C-band were concatenated to obtain a larger frequency coverage. With λ2 ranging from 1.8 × 10−3 m2 to 0.023 m2, the theoretical Faraday-depth resolution is δϕ = 165 rad m−2, with maximum observable Faraday depth ϕmax = 10123 rad m−2, and sensitivity to Faraday structures up to a scale of ∆ϕmax−scale = 1720 rad m−2. The RM CLEAN algorithm was used for deconvolution, and the RMTools package was applied for all RM synthesis.

The Stokes I images at three resolutions reveal the thin disk of NGC 891 and allow identification of compact emission features along the midplane. The lower-resolution maps are more sensitive to diffuse halo emission, extending several kpc above and below the disk plane. The 5″ resolution image was used by V. Heesen et al. (2025), who derived a scale height of 0.13 ± 0.01 kpc for the thin disk and 1.00 ± 0.09 kpc for the thick disk. The radial extent of the continuum emission closely resembles the radial extent of the bright H II regions and the brighter stellar disk. The inner disk is notably brighter in the north-east. At larger heights the emission becomes more diffuse and exhibits protrusions in the far outer halo.

The three-dimensional RM synthesis and RM clean procedures revealed that most spatially extended polarized emission is confined to Faraday depths within ±150 rad m−2. The first robust detection of polarized emission occurs at the Faraday depth of −630 rad m−2, while the last appears at +365 rad m−2. A polarized background radio source is detected through the halo of NGC 891 at RA = 02h 22m 44.87s and DEC = +42° 24′ 54.54″ with a Faraday depth of −176.2 ± 2.5 rad m−2.

A localized region in the north-east side of the galaxy shows an enhancement in polarized intensity (not in percentage polarization). The clean FDF spectrum at RA = 02h 22m 34.95s and DEC = +42° 21′ 41.20″ shows three peaks exceeding the detection threshold, located at Faraday depths of −630, −45, and +365 rad m−2. This region coincides with bright diffuse extraplanar Hα emission, including filaments Fil-WF3-3/4 and Fil-WF3-5. The neighboring peaks are neither symmetric nor mirror images of the sidelobe structure expected from the RMTF, suggesting they represent distinct Faraday-rotating components along the line of sight. The complex Faraday depth spectrum overlaps with H II regions 12-14.

Another notable region exhibiting complex Faraday rotation exists near the superbubble WF3-1SS1 (RA = 02h 22m 33.602s, DEC = +42° 21′ 10.20″), revealing two peaks above the detection threshold at Faraday depths of ϕ = −55 rad m−2 and ϕ = 280 rad m−2. At ϕ = 280 rad m−2, the region itself exhibits polarized emission, while the surrounding areas appear devoid of significant signal. In contrast, at ϕ = 120 rad m−2, the opposite is observed: polarized emission is detected in the surrounding area, but not within the target location.

The percentage polarization map was derived using the synchrotron-only Stokes I cube, after subtraction of the thermal emission contribution. The polarized intensity map was constructed by selecting, for each pixel, the maximum PI value along the Faraday depth cube that exceeded the detection threshold of 8σ. The rotation measure (RM) map is a peak-RM map, corrected for foreground Galactic Faraday rotation of −55.7 ± 20 rad m−2.

The magnetic field structure of NGC 891 shows that the plane-of-sky magnetic field orientation corrected for Faraday rotation does not show a magnetic field predominantly oriented along the galactic plane. The pattern closely resembles the magnetic field structure derived by S. Sukumar & R. J. Allen (1991) at 6.2 cm. The magnetic field in the plane of the sky has a significant component perpendicular to the disk, leaving no compelling alignment with the major axis across most of NGC 891. The overall structure resembles the X-shaped magnetic field structure reported by M. Krause (2008) and M. Krause et al. (2020).

The averaged RM profile along the projected major axis shows a smooth variation of RM along the major axis, with nearly constant RM on either side of the center and a resolved transition region near the projected minor axis. The interpretation of this RM structure depends critically on the true Galactic foreground Faraday rotation. Only if a sign change in RM occurs near the apparent minor axis, can this RM structure be interpreted in terms of a large-scale axially symmetric magnetic field. With the revised foreground of −55.7 ± 20 rad m−2, there remains the possibility of an RM sign change across the minor axis, although the Galactic foreground needs to be determined with higher precision for a definitive conclusion.

The pattern observed in the percentage polarization map is consistent with what has been reported in other edge-on galaxies: the thin disk exhibits a lower degree of polarization, whereas the halo shows higher fractional polarization. The depolarization analysis at nine representative locations shows that wavelength-dependent depolarization is at most a factor ∼ 10 in p2λ, which corresponds to a depolarization factor ∼ 1/3 in fractional polarization over the λ2 range of C-band and S-band combined. The data were fitted with the internal Faraday dispersion model of D. Sokoloff et al. (1998), yielding σϕ values ranging from approximately 33 to 123 rad m−2.

The depolarization in the galactic plane shows modest wavelength dependence with weak wavelength dependence in the halo. This is remarkable because the density of the warm ionized medium is expected to decrease significantly over the vertical distance probed. The weak wavelength dependence in the disk and in the halo is an important clue to the nature of the depolarization. The data suggest that σϕ λ2 ≲ 1, indicating that the diffuse polarized emission does not probe the entire disk, in analogy to the polarization horizon in the Milky Way. Both estimates for the depth of the layer from which detectable polarization is received indicate that we are observing a region with depth that is a considerable fraction of the radius of the radio disk.

The spatial distribution of H II regions in the disk of NGC 891 was investigated using the Hα spectral cube. A polar map of the galaxy was constructed by converting the positions of the detected H II regions into galactocentric coordinates, assuming circular orbits with a constant rotational velocity of 221.59 km s−1. The face-on map reveals a noticeable asymmetry in the radial distribution of H II regions between the two sides of the galaxy. On the north-east side, H II regions 15-24 are located closer to the galactic center, whereas those on the south-west side are distributed farther out. This asymmetry is consistent with the trailing spiral structure of NGC 891, although a fit of a logarithmic spiral to the H II regions suggested these H II regions are not all on the same spiral arm.

An intriguing feature revealed by the Hα analysis is the presence of an H II region located beyond the visible star-forming disk of NGC 891, at a galactocentric radius of 16.9 kpc, with both Hα and far-UV counterparts, indicating star formation outside the thin disk. The presence of both Hα and UV emission at such a large galactocentric radius implies that the star-forming disk of NGC 891 may extend farther than previously assumed.

The most significant asymmetry in the halo is the sudden broadening of the Stokes I contours in the north-east side coinciding with tracers of hot massive stars in the disk. The present work indicates that this is a localized feature in the outer disk of NGC 891. The notion of a region of localized enhanced star formation is also supported by the coincidence of significant diffuse X-ray emission that is otherwise only seen in the central region of NGC 891. The highest polarized intensity occurs at major axis offsets of −81.5″ and −95″, where there is also higher nonthermal total intensity at 3 GHz and at 6 GHz compared to equivalent positions in the southwest. The excess total intensity at 3 GHz relative to the corresponding point in the southwest is 523 µJy Beam−1 and 768 µJy Beam−1 for the two profiles, with percentage polarization of 33% and 26% respectively.

J. Rossa et al. (2004) identified a supergiant shell WF3-3-SS2 and a supershell WF4-SS1 with H II regions 15-20. These supershells have dimensions of 700 × 511 pc (WF3-3-SS2) and 345 × 244 pc (WF4-SS-1). Both these supershells are contained within the ∼ 2 kpc region with enhanced polarized intensity. The interior of such a supershell contains low-density hot, ionized gas visible in soft X-rays, conditions that would minimize the effects of Faraday rotation and thereby reduce depolarization, allowing polarized emission to remain clearly detectable. The morphology and multiwavelength properties suggest a large superbubble powered by a relatively young star cluster.

Most locations show a single peak in the Faraday depth spectrum, with the spread of the RM Clean components indicating a Faraday depth range ≲ 100 rad m−2. In a few locations, multiple Faraday depth components are resolved. The triple peak in the FDF spectrum near H II regions 12-14 requires a complex configuration. The two extreme Faraday depth components are slightly offset with respect to each other, between Hα filaments that appear to indicate an outflow from H II regions 12-14. The interpretation involves an asymmetric shell with density enhancement on one side, for example resulting from a density gradient, that can act as a Faraday screen creating two Faraday depth components with a spatial displacement comparable to the diameter of the shell. This model requires an asymmetric superbubble and the observer's line of sight at a large angle with the large-scale magnetic field.

The paper concludes with the following key findings:

1. A region on the north-east side of the galaxy shows high polarized intensity, accompanied by diffuse X-ray emission and a local excess in Stokes I. This region was previously found to contain multiple superbubbles, and it appears on the Earth-facing side of NGC 891, near the edge of the star-forming/radio continuum disk. The morphology and multiwavelength properties suggest a large superbubble powered by a relatively young star cluster.

2. The RM map of NGC 891, corrected for foreground Faraday rotation, shows a smooth transition along the major axis, between the north-east (mean RM = 4.50 rad m−2) and the south-west (mean RM = 48.42 rad m−2). A compact region of high positive RM (∼ 250 rad m−2) toward the galaxy's center is identified.

3. Across both disk and halo, the fractional polarization decreases toward the midplane but varies only weakly with wavelength. This indicates that the region from which polarized radio emission is detected has small internal Faraday dispersion and small differential Faraday rotation within the beam from any plasma in the foreground. Most of the detected polarized emission likely originates on the Earth-facing side of NGC 891. The low fractional polarization in the midplane is therefore primarily due to unpolarized synchrotron emission.

4. Most of the bright H II regions in the northeast side of NGC 891 are closer to the center than the H II regions in the southwest. The H II regions on both sides of the galaxy probably do not follow a common logarithmic spiral arm. In addition, a faint, isolated H II region was identified at a galactocentric radius of 16.90 kpc with Hα and far-UV counterparts, indicating recent star formation well outside the main star forming disk.

The paper notes that future work will extend the analysis to include L-band data, which will improve the resolution in Faraday depth, though this will be more useful only in selected areas where significant polarized intensity exists at these longer wavelengths.

Improvements to AI Systems Based on This Paper:

1. AI for Faraday Tomography and RM Synthesis Automation

• Train an AI model to automatically perform RM synthesis, RM CLEAN, and Faraday depth spectrum decomposition on radio polarization data cubes.

• The improved system can detect multiple Faraday components along a line of sight, separate physical components from sidelobe artifacts, and flag complex regions (e.g., triple peaks near H II regions) for follow-up analysis.

2. AI for Multi-Wavelength Source Association and Classification

• Develop a multimodal AI that cross-correlates radio continuum, polarization, Hα, far-UV, and X-ray data to identify star-forming regions, superbubbles, and outflows.

• The system can automatically detect isolated H II regions beyond the optical disk (like the one at 16.9 kpc) and classify them as genuine star formation sites versus background sources.

3. AI for Magnetic Field Structure Reconstruction

• Implement a deep learning model that takes Faraday depth cubes and polarization angle maps to reconstruct 3D magnetic field geometries in edge-on galaxies.

• The improved system can distinguish between X-shaped halo fields, plane-parallel fields, and asymmetric configurations, and predict whether observed RM gradients imply a sign change across the minor axis.

4. AI for Depolarization Modeling and Physical Interpretation

• Create an AI that fits wavelength-dependent depolarization models (e.g., internal Faraday dispersion, differential Faraday rotation) to multi-frequency polarization data.

• The system can automatically estimate Faraday dispersion (σϕ) and infer the vertical extent of the magneto-ionic medium, distinguishing between beam depolarization and intrinsic source effects.

5. AI for Foreground Galactic Faraday Rotation Correction

• Build a predictive model that uses Galactic all-sky RM maps, Hα emission, and dust templates to estimate foreground RM toward external galaxies with uncertainties.

• The improved system can provide robust foreground subtraction (e.g., refining the −55.7 ± 20 rad m−2 value) and flag cases where the sign of intrinsic RM remains ambiguous.

6. AI for Asymmetry and Outflow Detection in Disk-Halo Interfaces

• Train a convolutional neural network on radio continuum, polarization, and X-ray images to automatically identify localized enhancements (like the NE superbubble region) and trace their connection to star formation.

• The system can quantify scale heights, detect protrusions in the halo, and correlate them with Hα filaments and X-ray cavities to infer feedback processes.

7. AI for Synthetic Observation Planning and Sensitivity Forecasting

• Use the observational parameters (frequency coverage, uv-tapering, resolution) to train a generative model that predicts the expected RM fidelity and polarization detectability for given telescope configurations.

• The improved system can optimize future VLA or SKA observations by suggesting frequency ranges, integration times, and tapering schemes to resolve specific Faraday depth structures.

8. AI for Automated Physical Parameter Inference from Polarized Intensity Maps

• Develop a Bayesian neural network that ingests polarized intensity, RM, and depolarization maps to infer physical quantities like thermal electron density, magnetic field strength, and star formation rate.

• The system can output uncertainty estimates for each pixel, enabling robust comparisons across galaxies and cosmic time.

9. AI for Cross-Scale Feature Matching in Edge-On Galaxies

• Implement a graph neural network that links compact H II regions, supershells, and Faraday depth components across spatial scales (from 5″ to 20″ resolution).

• The improved system can automatically trace how a single superbubble produces multiple RM components (as seen near WF3-1SS1) and predict where polarized emission will be suppressed or enhanced.

10. AI for Anomaly Detection in Polarization Data

• Train an unsupervised anomaly detector on Stokes Q, U, and I cubes to find rare features like the background polarized source at ϕ = −176 rad m−2 or unexpected Faraday depth peaks.

• The system can flag these for human review, enabling discovery of background sources, intervening Faraday screens, or new physical processes in galactic halos.