Exploring the Dust Content of Galactic Halos with Herschel III. NGC 891

Spitzer images at 3.6 ${\mu}$m, 4.5 ${\mu}$m, and 24 ${\mu}$m are used in this study. Images taken with the Infrared Array Camera (IRAC) at 3.6 ${\mu}$m and 4.5 ${\mu}$m cover the entire galaxy (PID: 3, PI: G. Fazio). The observations consist of a series of dithered images taken with exposure times from 0.4 to 96.80 seconds. The basic calibrated data were obtained from the Spitzer archive and combined into a mosaic using the montage package (http://montage.ipac.caltech.edu/) correcting for variations in the overall level and rejecting outliers. The point source flux accuracy is $\sim$2% (Reach et al., 2005) and the extended source surface brightness accuracy is estimated at $\sim$4%.

We observed NGC 891 with the Multiband-Imaging Photometer for Spitzer (MIPS, Rieke et al., 2004) at 24 ${\mu}$m (PID: 20528, PI: C. Martin). The images were obtained using the medium rate, scan-map mode and the scan legs of $0.5^{\circ}$ with 148″ cross-scan offsets. This observation consists of 5 scans (Obs. id 14815488, 14815744, 14816000, 14816256, 14816512) with total exposure times of 1600 seconds. The data were reduced with the MIPS Data Analysis Tool v3.04 (DAT Gordon et al., 2005). Extra steps were carried out to improve the images including readout offset correction, array averaged background subtraction (using a low order polynomial fit to each leg, with the region including NGC 891 excluded from this fit), and exclusion of the first five images in each scan leg due to boost frame transients. The point source flux accuracy is $\sim$2% (Engelbracht et al., 2007) and the extended source surface brightness accuracy is estimated at $\sim$4%.

The 24 ${\mu}$m image resolves stars and background galaxies, which we mask out for accurate photometry of the galaxy. We identified background galaxies using SExtractor. Briefly, we set the detection threshold to 1.5$\sigma$ in surface brightness. The sources identified by the segmentation map were then replaced by the values in the background map at the same locations.

The Photodetector Array Camera and Spectrometer (PACS, Poglitsch et al., 2010) observation for NGC 891 was conducted as our cycle 1 open-time program (OT1_sveilleu_2, PI: S. Veilleux). This program utilizes PACS blue 70 ${\mu}$m and red 160 ${\mu}$m channels with 6 scan position angles at 55^∘, 70^∘, 85^∘, 95^∘, 110^∘, and 125^∘ (Obs IDs: 1342237999, 1342238000, 1342238001, 1342238002, 1342238003, and 1342238004). The exposure time per each scan was 5596 seconds (total 9.3 hours) including overhead. The scan map leg length and separation were 4′ and 4″ and the number of scan map legs was 60. Each scan map was repeated three times for redundancy reasons.

We also retrieved, reduced, and analyzed the PACS data from The Herschel EDGE-on galaxy Survey (HEDGES, Murphy, 2011, OT2_emurph01_3) for the analysis of extended structure as these data have a larger field-of-view than do our observations. For the blue channel (70 ${\mu}$m), the scan angle was 45^∘ and 135^∘ with an exposure time of 6550 seconds. The scan map leg length and separation were 14$\aas@@fstack{\prime}$6 and 14$\aas@@fstack{\prime\prime}$0 and the number of scan map legs was 70. For the green channel (100 ${\mu}$m), the exposure time was 8977 seconds at scan angles of 45^∘ and 135^∘. The scan map leg length and separation were 14$\aas@@fstack{\prime}$6 and 20$\aas@@fstack{\prime\prime}$0 and the number of scan map legs was 51.

The observation in 250 ${\mu}$m, 350 ${\mu}$m, and 500 ${\mu}$m bands were performed using the Spectral and Photometric Imaging Receiver (SPIRE, Griffin et al., 2010) onboard Herschel Space Observatory. Our SPIRE images are the part of the KINGFISH open-time key program, KPGT_cwilso01_1, from Herschel guaranteed time key project,Physical Processes in the Interstellar Medium of Very Nearby Galaxies¹¹1http://hedam.lam.fr/VNGS/index.php (ID: 1342189430, Griffin et al., 2010). The beam sizes are 465.39, 822.58, and 1768.66 arcsec² at 250, 350, and 500 ${\mu}$m, respectively (SPIRE Data Reduction Guide, Table 6.7) and the exposure times were 1678 seconds for each channel.

For the PACS data, we used the Herschel Interactive Processing Environment (HIPE, Ott, 2010) version 10.0 for data reduction. This reduction procedure contains the extraction of the calibration tree for the data processing, electronic crosstalk correction, flat-fielding, and unit conversion from volts to Janskys per pixel. The final map was created with the algorithm implemented in Scanamorphos (v24.0, Roussel, 2013). The error and weight maps were also produced with Scanamorphos and the error map is defined as the error on the mean brightness in each pixel.

The SPIRE data were processed from level 0 up to level 1 with HIPE scripts in the Scanamorphos distribution. The same steps as PACS processing were conducted. The thermal drifts are subtracted by using the smoothed series of thermistors located on the detector array as the input of the drift model (see Ott (2010) for details). A more detailed description of the data reduction can be found in Paper I.

Both sets of PACS observations contain a low-amplitude artifact parallel to the major axis on the eastern side of the disk midplane. It is 70″ away to the south-east and the flux in the ghosts is less than 0.5% of the corresponding object flux. This region immediately east of the disk midplane was not used in our analysis of the vertical structure of the disk. The amplitude of the feature is small enough, that it has a negligible effect on our integrated flux measurements.

Figure 1 compares the new (24, 70, 160 ${\mu}$m) images to the SPIRE images at 250, 350, and 500 ${\mu}$m. We define the center of NGC 891 by the location of the maximum 4.5 ${\mu}$m IRAC intensity, a proxy for the center of the stellar bulge. In Figure 1 and subsequent images, the coordinates (0,0) reference this position. Table 1 provides the corresponding sky coordinates.

We measured fluxes from the 24, 70, 160, 250, 350, and 500 ${\mu}$m images. In order to match the point spread function (PSF), we convolved the images with the kernels provided by Aniano et al. (2011). The image pixel scale and coordinates were then matched using the IDL code hrebin.pro and hastrom.pro in the Goddard package. These processes conserve total image flux.

The aperture adopted for the integrated photometry is based on a growth curve measured from photometry of the SPIRE 500 ${\mu}$m image in increasingly larger apertures. It is an elliptical aperture of semi-major axis of 7$\aas@@fstack{\prime}$35 and semi-minor axis of 3$\aas@@fstack{\prime}$68 that includes most of the dust emission seen in 24 ${\mu}$m image after convolution to the PSF of the 500 ${\mu}$m image. Figure 2 shows this aperture on the 24 ${\mu}$m image.

Within an annulus at larger radii, we masked point sources and background galaxies based on the morphology in the 24 ${\mu}$m data. The sky background was taken as the mean data value following 2$\sigma$ clipping. The variance in this region was used to compute the standard deviation of the mean. This uncertainty on the mean background dominates the total background error for large apertures. We also include a term for the pixel-to-pixel statistical variance in the background.

The error budget is dominated by the calibration uncertainties, which exceed the uncertainties from both background subtraction and photon noise. Following Paper I, we conservatively estimate the background error at 10% in all bands.

We computed color corrections for the PACS and SPIRE fluxes assuming a blackbody spectrum modified by the dust emissivity, $\kappa_{\nu}=\kappa_{\rm 0}(\nu/\nu_{\rm 0})^{\beta}$ with $\beta=2$, and a blackbody temperature of $T=30$ K.²²2 As the PACS and SPIRE pipelines compute fluxes assuming a monochromatic spectrum, we applied a color correction. See section 4.3 in “PACS Photometer Passbands and Color Correction Factors for Various Source SEDs”, PICC-ME-TN-038, version 1.0, and Table 5.7 in SPIRE Handbook, HERSCHEL-DOC-0798, version 2.5. We divided the original flux measurements by the following correction factors: 0.977, 1.037, 0.9796, 0.9697, and 0.9796 for, respectively, 70 ${\mu}$m, 160 ${\mu}$m, 250 ${\mu}$m, 350 ${\mu}$m, and 500 ${\mu}$m. The estimated aperture corrections were small compared to the photometric errors and were therefore not applied.

The integrated fluxes listed in Table 1 for the 24, 70, 160, 250, 350, and 500 ${\mu}$m bands. Our measurements agree with both Bendo et al. (2012) and Hughes et al. (2014) to within the $1\sigma$ errors.

The arrow in Fig. 2 indicates a filament of 24 ${\mu}$m emission extending 6$\aas@@fstack{\prime\prime}$90 (19.6 kpc) southeast of the dust halo. We discovered this filament in the Spitzer MIPS data and confirmed it using Herschel. Figure 1 shows the SPIRE detections. The structure is outside the field-of-view of our new PACS images, so we examined the PACS images obtained previously by The Herschel EDGE-on galaxy Survey (HEDGES, Murphy, 2011) and found a clear detection at at 160 ${\mu}$m, a few clumps of emission at 100 ${\mu}$m, and weaker emission at 70 ${\mu}$m. The far-infrared filament is therefore not an instrumental artifact.

We looked for a background cluster of galaxies at this location but found none in either X-ray images or optical images.³³3The $z=0.0184$ cluster Abell 347 lies to the southeast at 02:25:50.9,+41:52:30 (NED). Our photometry indicates a broad spectral energy distribution, which we show is consistent with thermal dust emission in Section 4. The far-infrared filament may therefore be a component of the dusty halo of NGC 891, and we will refer to this morphological feature as the dust spur. Spectroscopy of an emission line would confirm its association with NGC 891; alternatively, higher-resolution mapping would be useful to definitively rule out an unresolved galaxy cluster.

The morphology of the dust spur shows no connection to the center of the galaxy, making an association with a galactic outflow unlikely. When projected on the sky, the long-axis of the dust spur is roughly perpendicular to the disk. The filament emerges at a galactocentric radius $R$ = 4$\aas@@fstack{\prime}$0 (11 kpc) and extends 6$\aas@@fstack{\prime}$9 (19.6 kpc) to the southeast, and has a width of $\sim$ 2$\aas@@fstack{\prime}$0 (6 kpc).

The dust spur is not coincident with the large filament of neutral hydrogen described by Oosterloo et al. (2007) (see also Pingel et al., 2018). In Figure 3, we show their H I 21-cm emission contours on the Herschel and Spitzer images. The lowest contour at $N(HI)=1\times 10^{19}\rm~{}cm^{-2}$ extends 22 kpc northwest of NGC 891. The dust spur extends to the southeast, in contrast, and is not visible in the H I contours. We will show that the H I filament and the dust spur have different compositions in Section 4.

The middle panel of Fig. 3 compares the dust emission to the radio continuum. The overall shape of the radio continuum contours is similar to the distribution of far-infrared emission, whereas the H I contours show a lower ratio of vertical height to galactocentric radius.

A close inspection of the dust spur suggests a spatial coincidence between the brightest emission regions at 250 ${\mu}$m and the local maxima in the the faintest radio continuum contours. This correlation should be interpreted cautiously. Unresolved sources in the SPIRE map could be background sources unrelated to NGC 891, and the radio detections have low signal-to-noise. If the radio sources are at the redshift of the dust spur, then the radio flux levels of the clumps, 0.2 mJy beam^-1, combined with the fluxes listed in Table 3, are consistent with far-IR – radio relation defined by star-forming regions (Yun et al., 2001; Bell, 2003). These data raise the possibility that the dust spur is a region of active star formation.

We smoothed the 70  ${\mu}$m image to match the resolution of the 160  ${\mu}$m image, and then we divided the smoothed 70  ${\mu}$m image by the 160  ${\mu}$m image to produce a color map from the PACS data. This color map resolves a broad plume of thermal dust emission, indicated by the arrow in Figure 4. This structure emerges from a region of the disk within roughly 1$\aas@@fstack{\prime}$8 (5.1 kpc) of the nucleus, slightly inside the 3$\aas@@fstack{\prime}$1 radius of the molecular ring. It extends northwest, roughly perpendicular to the disk, reaching a height of 2$\aas@@fstack{\prime}$7 (7.7 kpc). We refer to this structure as the superbubble but will argue in Section 5 that a galactic wind is actually developing here. We will show that this region has a slightly higher dust temperature than the surrounding halo in Section 4.

Contours of the soft X-ray emission (Hodges-Kluck & Bregman, 2012) outline the perimeter of the dusty superbubble. In deeper X-ray observations, this hot ($kT=0.71\pm 0.01$ keV) gas is concentrated near the most active region of star formation and coexists with gas near the virial temperature, $kT=0.20\pm 0.01$ keV (Hodges-Kluck et al., 2018). Harder diffuse X-ray emission is detected within 5 kpc of the galactic center; its physical origin remains unclear.

The soft X-ray emission from galactic winds is produced mostly at the interface between the hot wind and a cooler component of the multi-phase outflow (Strickland et al., 2004a, b). The smaller panels in Figure 4 overlay the radio continuum contours and the H$\alpha$ contours on the PACS color map. The H$\alpha$ image shows one prominent filament extending well into the dusty bubble, but the H$\alpha$ image is not sensitive enough to determine the amount of correlation with the superbubble morphology at other wavelengths. We have shown in Figure 3 that the radio contours closely follow the morphology of the dusty halo. As would therefore be expected, the radio contours do not describe the PACS color map well.

The fate of the superbubble will be determined in large part by the CGM. The inner CGM of NGC 891 contains much more cold gas than hot, virialized gas (Hodges-Kluck et al., 2018; Das et al., 2020)

On the eastern side of the disk, the 24 ${\mu}$m emission extends 2$\aas@@fstack{\prime}$03 (5.76 kpc) over most of the disk, reaching a larger distance along the dust spur (Fig. 3). On the western side of the NGC 891 disk, the dusty halo is detected out to 3$\aas@@fstack{\prime}$68 (10.5 kpc) in the 24 ${\mu}$m map.

The extent of the PAH emission is similar to that of the cold dust. McCormick et al. (2013) estimate a symmetric extent of $\pm 5\aas@@fstack{\prime}0$ with a bulk height, $H_{e,PAH}=7.1$ kpc.

The top panel of Figure 6 shows the SED for the entire galaxy. We fit only the 70 $-$ 500 ${\mu}$m data because the 24 ${\mu}$m flux comes from a warmer dust component. The single-temperature, MBB model with $\beta\equiv 2$ gives a dust mass of $M_{d}=1.1\times 10^{8}$${\rm~{}M_{\odot}}$ and dust temperature $T=21.7\pm 0.2$ K. We also plot the fit with $\beta$ taken as a free parameter. A comparison shows that our fitted dust parameters are robust to minor changes in the dust emissivity; see Table 2 for a quantitative comparison.

The dominant error term is the 10% uncertainty in the flux calibration. To determine the uncertainty on the fitted mass and temperture, we scale the photometry by 10% and refit the models. We find that systematically increasing (or decreasing) all the fluxes can increase (decrease) the fitted dust masses by $\sim$10-20%, but dust temperature varies by less than a degree. Our results confirm the dust mass and temperature obtained previously by Hughes et al. (2014); Bocchio et al. (2016).

The total dust mass fit to the FIR observations is nearly twice that inferred from radiative transfer modeling of the distribution of optical and near-IR emission. For example, when scaled to our adopted distance, Xilouris et al. (1998) found $5.3\times 10^{7}$${\rm~{}M_{\odot}}$ of dust. A significant fraction of the dust presumably goes undetected because little optical/near-IR radiation emerges from some regions of the galaxy. Previous work has suggested that this hidden dust could reside in a geometrically thinner dust disk (Popescu et al., 2000, 2011) or in clumpy clouds associated with molecular gas (Bianchi, 2008). Seon et al. (2014) considered NGC 891 specifically and introduced a thin dust disk of scale height $\approx 5\arcsec$ to simultaneously describe the distribution of GALEX UV emission (Morrissey et al., 2007) and Herschel SPIRE photometry (Bianchi & Xilouris, 2011).

For completeness, we point out that the impact of smaller dust grains (grain radii $<50$ Å) on our fits is constrained by the 24 ${\mu}$m flux. We fit a two temperature model to the full SED including the 24 ${\mu}$m flux. The temperature of a warmer component is not well constrained by this single data point, so we assumed a dust temperature of $T=220$ K to facilitate comparison to NGC 4631. In Paper I, we found an additional dust component at 220 K in NGC 4631. In their single-temperature, MBB fits, they treated the 70 ${\mu}$m flux as an upper limit because the contribution from the warmer component was non-negligible. We find, however, that the contribution of this warmer component to the 70 ${\mu}$m flux is only 0.3% for this two temperature model fitted to NGC 891. Assuming that the peak wavelength of the MBB curve for the warmer component is shorter than 24 ${\mu}$m, we vary the temperature of the warmer component, and the contribution to 70 ${\mu}$m flux is 1.3% at most. Therefore, we consider 70 ${\mu}$m data point in the same way as other wavelength data in the fitting process. For an assumed warm dust temperature of 220 K, we fit a two-component model and find dust mass in the 220 K component is negligible ($\sim 0.00003$%) compared with dust mass in the $\sim 22$ K component. The mass in the cool component in this two-component fitting does not deviate from the mass in the one-component, cool component only, fitting as shown in Table 2.

Table 2 summarizes the dust properties for the regions labeled in Figure 5. The photometry shown in Figure 6 was obtained after degrading the resolution of the 70, 160, 250, and 350 ${\mu}$m maps to match that of the 500 ${\mu}$m map; see details in Section 2.

Our analysis constrains the contribution from dust well beyond the galactic plane where the starlight is too faint to model its attenuation. Inspection of Table 2 indicates 87% of the dust mass is associated with the galactic disk (log ($M_{\rm dust}/M_{\odot}$)$=7.970\pm 0.006$) with 65% of the disk dust concentrated in the inner disk (log ($M_{\rm dust}/M_{\odot}$)$=7.781\pm 0.012$). We associate roughly 13% of the total dust mass with the halo component at $|z|>0\aas@@fstack{\prime}5$, or $\pm\ 1.42$ kpc; the same halo aperture includes 20% of the total H I mass. This difference can be attributed to the lower dust-to-gas ratio of the halo. Roughly 25% of the halo dust mass (log ($M_{\rm dust}/M_{\odot}$)$=7.163\pm 0.053$) comes from the superbubble region, while the dust spur, log ($M_{\rm dust}/M_{\odot}$)$=5.88\pm 0.02$, makes a smaller contribution.

The halo dust temperature ($T=19.24\pm 0.62$ K) is significantly cooler than the disk dust ($T=21.89\pm 0.06$ K). The superbubble dust ($T=20.61\pm 0.03$ K) is warmer than the surrounding halo but cooler than the disk. The emission from the dust spur is weak at 70 ${\mu}$m compared to the SPIRE bands, and the fitted dust temperature is considerably lower than that of the disk or halo.

We also examined the variation in dust properties at the resolution limit of the 500 ${\mu}$m image. We fit the SED of all pixels that had $S/N>3\sigma$ in all five bands, essentially the entire disk component and the disk – halo interface. Hughes et al. (2014) previously found the variation in dust temperature to be more uneven than that of the dust mass. Figure 7 shows our resulting maps of the temperature and dust surface density, i.e., the column density of dust in mass units.

The dust surface density decreases with distance from the nucleus along the major axis. In the vertical direction, the dust surface density drops very quickly to values less than 0.50${\rm~{}M_{\odot}}$ pc^-2.

The temperature range in Fig. 8 is similar to the $T_{d}=17-24$ K range found previously by Hughes et al. (2014). The hotspots identified in both the 24 ${\mu}$m map and the 70 ${\mu}$m/160 ${\mu}$m color map stand out as regions with $T_{d}\geq\ 22.5$ K in Figure 7. This region of the disk is surrounded by a molecular ring (Israel et al., 1999; Scoville et al., 1993). The dust cools off with increasing radius in the disk beyond the molecular ring.

We compare the dust temperature and surface density directly to the 3.6 ${\mu}$m and 24 ${\mu}$m emission in Figure 8. The 24 ${\mu}$m emission is a good proxy for the star formation rate surface density (Calzetti et al., 2007), while the 3.6 ${\mu}$m emission traces the stellar mass density. We have color coded the points in Figure 8 by their location in NGC 891.

The dust in the midplane of the inner disk has temperature, $T_{D}\approx 22-23$ K and a high mass column, as indicated by its location in the upper right of Figure 8. With increasing vertical height at fixed radius, the temperaure decreases by roughly one degree as the surface brightness drops by an order of magnitude. This locus (blue points) defines a narrow band that continues smoothly into the emission from the superbubble region where the temperature drops a few more degrees (orange and yellow points). Hughes et al. (2014) produced color maps that identify this same spectral transition with height above the disk. Adding the halo emission immediately above the inner disk extends this correlation to dust temperatures $T_{D}\lower 2.0pt\hbox{$\buildrel{\scriptstyle<}\over{\scriptstyle\sim}$}20$ K. Based on this correlation between dust temperature and 24 ${\mu}$m emission, the dust in and above the inner disk is heated primarily by the starlight from the star-forming regions.

The width of the SB(24 ${\mu}$m) – $T_{D}$ locus in the inner disk correlates with radial distance from the nucleus. For example, following the midplane points in Figure 8 from the inner disk to the outer disk (green points) follows the upper edge of the triangular locus. At fixed 24 ${\mu}$m surface brightness, dust in the midplane of the outer disk is cooler than dust above the plane at small radii. Equivalently, at fixed dust temperature, the superbubble and halo emission are not as bright at 24 ${\mu}$m as the outer disk. We expect a transition in the outer disk toward heating by an older stellar population, and it seems plausible this transition manifests as the broad triangle of green points in Figure 8. Consistent with this interpretation, the 3.6 ${\mu}$m emission from the midplane points in the outer disk (green points with cyan dot) are strongly correlated with the dust temperature. Dust above the midplane of the outer disk fills the interior of the triangular locus, as might be expected for dust heated by a mixture of the two stellar populations.

We acknowledge that the temperature correlations look nearly identical against the near-infrared (3.6 ${\mu}$m) and mid-infrared (24 ${\mu}$m) emission, so the temperatures have the same correlations with the tracers of stellar mass and star formation, respectively. It seems likely that this happens because the mass and SFR both decrease with separation from the center of the galaxy.

The panels on the right side of Figure 8 show power law correlations between infrared emission and dust surface density. For the inner disk, the correlation is tighter in the mid-infrared (24 ${\mu}$m), more closely related to SFR surface density. In the outer disk, the 3.6 ${\mu}$m emission shows the stronger correlation, consistent with a closer relation to stellar mass surface density.

Table 2 compares the dust mass to the gas mass, where all measurements taken from the literature have been scaled to an NGC 891 distance of 9.77 Mpc. The total H I mass of NGC 891 is then $4.4\times 10^{9}$${\rm~{}M_{\odot}}$(Oosterloo et al., 2007). The total CO(1-0) intensity (Scoville et al., 1993; Garcia-Burillo et al., 1992) correspondes to a molecular gas mass of $M({\rm H_{2}})\approx 4.2\times 10^{9}\mbox{${\rm~{}M_{\odot}}$}$ for the Bolatto et al. (2013) calibration of the CO-to-$\rm H_{2}$ conversion factor, $X_{\rm CO}=2\times 10^{20}{\rm cm}^{-2}{\rm[K~{}km~{}s^{-1}]}^{-1}$. The ratio of molecular-to-atomic gas is close to unity, $M({\rm H_{2}})/M(\text{H\,{I}})\approx 0.95$, which is higher than the median of 0.6 found for Sab galaxies (Obreschkow & Rawlings, 2009). Including the mass contribution from helium, the total mass of cold gas is $1.2\times 10^{10}$${\rm~{}M_{\odot}}$.

Comparing our measured dust mass to the atomic and molecular gas mass, we find a global gas-to-dust ratio $M_{\rm gas}/M_{\rm dust}\approx 110$, where the uncertainty introduced by the $X_{\rm CO}$-factor is 30% (Bolatto et al., 2013). The gas-to-dust ratios for the SINGS galaxies range from 100 to 340 after scaling the values in Table 4 of Draine et al. (2007) to the same $X_{\rm CO}$ and including the helium mass. On this scale, the gas-to-dust ratio of the Milky Way is 200 based on the $M_{\rm dust}/M_{\rm H}\approx 0.007$ ratio given by Draine et al. (2007). Based on its atomic and molecular gas content, NGC 891 is dustier than the median Sab galaxy.

The molecular gas is concentrated in the inner disk of NGC 891, so it is not surprising that the inner disk is the dustiest region. The H I disk is larger than the molecular disk. Oosterloo et al. (2007) attribute slightly over 70% of the total H I to the disk, and our measurements of their H I map indicate the inner disk H I mass is $1.22\times 10^{9}$${\rm~{}M_{\odot}}$, slightly less than half the mass of the entire H I disk. The total gas mass of the inner (and full) disk is then $7.4\times 10^{9}$${\rm~{}M_{\odot}}$ ($9.8\times 10^{9}$${\rm~{}M_{\odot}}$). These disk masses do not include the small contribution from warm ionized gas (Rand et al., 1990; Dettmar, 1990), but they do include the mass contribution from helium, as described in the notes to Table 2.

The filament northwest of NGC 891 contains at least $1.6\times 10^{7}$${\rm~{}M_{\odot}}$ of H I(Oosterloo et al., 2007). No thermal dust emission is detected from the H I filament. The upper limit on the dust mass excludes gas-to-dust ratios similar to galactic disks. If the H I filament is accreting gas as Oosterloo et al. (2007) suggest, then the lack of dust emission favors low metallicity gas. Alternatively, if the H I filament is fountain gas, the timescale to recycle the disk gas must be long enough to destroy even large grains, which seems unlikely based on the large amount of dust that persists in the CGM (Ménard & Fukugita, 2012).

The properties of the dust spur southeast of NGC 891 contrast sharply with those of the H I filament. At the distance of NGC 891, the far-infrared filament contains nearly $8\times 10^{5}$${\rm~{}M_{\odot}}$ of dust. Based on the dust mass, this circumgalactic gas was once in a galaxy, either NGC 891 or a satellite. Using the H I map (Oosterloo et al., 2007), the H I mass in the dust spur is less than $2\times 10^{7}$${\rm~{}M_{\odot}}$. The exceptionally low ratio of neutral hydrogen gas to dust is a puzzle. Perhaps a background source will be found for the infrared emission. An alternative, however, is that most of the gas mass is either molecular or ionized.

The mass ratio of H I to dust in the inner CGM of NGC 891 is indistinguishable from the galactic disk. We find this surprising because the neutral fraction of halo gas is around 1% (Popping et al., 2009; Bland-Hawthorn et al., 2017). The dust was presumably made inside galaxies and transported into the halo via outflows or stripping. Gas clouds are normally destroyed by these processes unless special conditions are met, so we would not expect the galactic gas to remain neutral. The inner CGM of NGC 891 contains more neutral gas than the median COS (Cosmic Origins Spectrograph) Halos galaxy (Prochaska et al., 2017; Das et al., 2020). The disk-like ratio of H I to dust certainly suggests that the excess of neutral gas in the inner CGM of NGC 891 was removed from galaxies. We will discuss the extrapolated dust mass of the CGM in Section 5.3.

We found an extraplanar, dusty region that is significantly warmer than the dusty halo. We called it the dusty superbubble because its location is coincident with thermal X-ray emission ($kT=0.71$ keV) likely associated with an outflow. In Section 5.1.1 below, we show that the thermal energy of this superbubble is sufficient for the superbubble to punch through the disk and blowout into the halo. Blowout in NGC 891 is of particular interest because the SFR surface density is a factor of three below the commonly assumed threshold of 0.1 $~{}{\rm M_{\odot}}$ yr^-1kpc^-2. The fate of this outflow depends largely on the halo gas density profile, which is reasonably well constrained in NGC 891. We present this evolution in Section 5.1.2 and discuss its relationship to the recent detection of cosmic ray advection in the halo of NGC 891.

We have detected thermal emission from $1.5\times 10^{7}$${\rm~{}M_{\odot}}$ of cool dust in the inner halo of NGC 891, i.e., the region within the ellipse drawn in Figure 2. We estimate the total halo dust mass from an extrapolation of the halo surface brightness profile. We use the 24 ${\mu}$m profile on the west side of the disk and assume the halo extends vertically from a base of radius 20.9 kpc. We equate the central surface brightness of 6.0 MJy sr^-1 with a mass surface density of 0.20${\rm~{}M_{\odot}}$ pc^-2 based on Figure 7, arriving at a total halo dust mass of $4\times 10^{7}$${\rm~{}M_{\odot}}$.

With this extrapolation, the halo dust mass increases from 15% to roughly 43% of the dust mass in the thick disk. The estimated dust mass in the halo therefore approaches the cosmic ratio of halo dust to disk dust. Ménard & Fukugita (2012) measured the reddening of background quasars caused by dust in Mg II absorbers and found that these absorbers account for most of the dust inside the virial radii of galaxies and that the galaxy halos contain at least 52% as much dust as do galaxy disks. While we cannot rule out the possibility of substantially more halo dust in a cooler component at $T\sim 7~{}\rm K$, the similarity of these numbers suggests that the Herschel imaging has detected the main component of halo dust in NGC 891.

We have identified two structures, a dusty superbubble and a dust spur, that may pollute the CGM with dust. To add further insight into exactly how dust is transported into the CGM, we applied unsharp masking techniques to highlight the high spatial frequency structure in the dust emission. Regardless of the exact procedure used, the results in Figure 9 show dust filaments emanating from the hotspots associated with the molecular ring as well as from the central hotspot. These filaments are detected over the full diameter of the molecular ring and extend roughly 4 kpc from the midplane. These dust filaments appear to be associated with the feedback processes that are mixing dust into the halo; they are found over a larger region of the disk than the base of the superbubble. It should be possible (in future work) to compare the properties of such filaments directly to wind simulations that take the interactions of the bubbles driven by multiple star-forming regions into account (Tanner et al., 2016).

The halo dust clearly traces a fluid that is not pristine gas. While this material most likely originated in galaxies, the question remains whether it was primarily blown out of the disk by feedback processes or whether tidal streams and ram pressure stripping of satellites have contributed a substantial fraction of the mass. The newly discovered dust spur clearly shows that interactions with satellites do play some role. The mass in the dust spur is about 25% of that in the superbubble, however; and the filaments are also lifting dust about 4 kpc off the disk midplane. Hence our results favor a dominant role for stellar feedback in enriching the halo with dust.

To determine whether it is energetically plausible for feedback to lift most of the gas associated with the halo dust out of the disk, we conservatively estimate the total mass of gas lifted into the halo along with the dust. The dust-to-gas mass ratio of the galaxy as a whole is $M(dust)/M(gas)\approx 1/110$. We estimate $M(gas)\approx 1.7\times 10^{9}$${\rm~{}M_{\odot}}$ just above the disk and, very roughly for the entire halo, $M(gas)\approx 4\times 10^{9}$${\rm~{}M_{\odot}}$. A lower dust-to-gas ratio would yield more halo gas associated with the dust, but the average galactic value is most appropriate for testing the hypothesis that the dust was lifted out of the disk by feedback. We note that $4\times 10^{9}$${\rm~{}M_{\odot}}$ is a small fraction of the total mass found in the cool CGM of galaxies (Werk et al., 2014, $7-12\times 10^{10}~{}\mbox{${\rm~{}M_{\odot}}$}$).

Following Howk & Savage (1997), we assume an isothermal sheet model for the vertical distribution of light and mass in NGC 891. Based on the similarity of the rotation curves for NGC 891 and the Milky Way, we adopt the Milky Way midplane mass density $\rho_{0}$ (Bahcall, 1984) as a proxy for NGC 891; the mass scale height $z_{0}$ is taken from infrared imaging of NGC 891 (Aoki et al., 1991). The potential energy, $W$, of a cloud of mass $M_{\rm cl}$ a height $z$ above the midplane is then

The energy required to lift the $4\times 10^{9}$${\rm~{}M_{\odot}}$ of H I, $\rm H_{2}$, and dust in the superbubble to a height of 4.3 kpc is roughly

For a constant SFR of 4.8$~{}{\rm M_{\odot}}$ yr^-1, supernovae and stellar winds provide a mechanical power, $L_{\rm w}\approx 1.2\times 10^{42}$ ergs s^-1. Assigning an efficiency $\epsilon$ for transferring this energy to the cool gas and dust, the time required to enrich the halo with dust is roughly

The current rate of star formation in the disk provides enough feedback energy to lift the dust and associated gas into the halo in just a few rotational periods only if the feedback efficiency is very high. However, if we picture the CGM as continually recycling disk material over a longer timescale, then feedback can easily account for most of halo dust in NGC 891. For example, if the disk of NGC 891 was assembled at $z\lower 2.0pt\hbox{$\buildrel{\scriptstyle<}\over{\scriptstyle\sim}$}\ 2$, and the SFR has steadily declined over the last 10.3 Gyr, then the dust can be lifted into the halo with inefficient feedback characterized by $\epsilon\approx\ 0.03$.

It is interesting to compare the feedback in NGC 891 to that in NGC 4631, another edge-on galaxy which we recently studied with Herschel in Paper I. The galaxies NGC 4631 and NGC 891 have comparable H$\alpha$, X-ray, and radio luminosities. Linearly polarized radio continuum emission has been detected in the halos of both galaxies and shows that magnetic field lines reach into the halos of both galaxies (Hummel et al., 1991). The FIR to radio flux ratio for NGC 891 is $q_{60\mu m}=2.19$ in Table 3. The FIR luminosity at 60 ${\mu}$m is $9.64-9.70\mbox{${~{}\rm L_{\odot}}$}$ (Sanders et al., 2003; Surace et al., 2004) and radio continuum luminosity is $21.85-21.76~{}\rm W~{}m^{-2}~{}Hz^{-1}$ (Condon et al., 2002; Murphy et al., 2009). Thus, we estimate $q_{60\mu m}=2.03-2.19$ for NGC 4631. This ratio is a sensitive probe of age during a starburst phase (Bressan et al., 2002), and the comparable values for NGC 4631 and NGC 891 indicate both galaxies are in a post-starburst phase.

In spite of these similarities, the halos of the two galaxies show some significant differences. The density and scale height of the warm ionized gas are a bit larger in NGC 4631, and the magnetic field extends further into the halo of NGC 4631 (Hummel et al., 1991). The direction of the magnetic field lines is not well constrained for either galaxy due to poorly constrained models for Faraday rotation, but after correction the foreground Faraday rotation the field lines are quite different in the two galaxies. The magnetic field lines point radially outward from the disk of NGC 4631 while no overall orientation is visible in NGC 891 where the field appears to be drawn into the halo by more local events (especially on the south side of the disk).⁴⁴4 The radio continuum emission in Fig. 4 also shows an extension in total intensity away from the disk plane to the southwest (opposite the dust spur). The linear polarization of this southwest extension is larger than the polarization in any other region of NGC 891, and the polarized emission extends further into the halo here than it does above the superbubble (Hummel et al., 1991). Bulk motion has convected the magnetic field outwards, and the adiabatic losses of the electrons may explain the changes in spectral index reported by Hummel et al. (1991). Whether there is a direct connection between this inferred velocity gradient and the dust spur remains unclear. Overall, these results suggest the galactic wind is stronger in NGC 4631 than in NGC 891.

Stronger feedback in NGC 4631 would be expected based on its higher SFR surface density and gas fraction. The SFRSD of NGC 4631 is about 5 times larger than that of NGC 891 as estimated in Section 5.1. It is also clear that NGC 4631 has a higher gas fraction than NGC 891 based on its higher H I mass, lower dynamical mass, and later spiral type (de Vaucouleurs et al., 1991).

It is therefore quite interesting to compare the dust properties of these two halos. We find the measured dust mass of NGC 891 is 2.6 times higher than NGC 4631 in Paper I. We attribute this result to the larger size and mass of NGC 891. The superbubble in NGC 4631 is slightly smaller and contains a factor of two less dust mass than the NGC 891 superbubble.

The temperature of the dust, however, differs significantly between the two superbubbles. The temperature of the dusty superbubble is significantly higher in NGC 4631, $T_{\rm dust}=24.46\pm 0.77$ K. When we include the 100 ${\mu}$m data point from the HEDGES data, the superbubble temperature for NGC 891 increases by $\sim 1$ K which is still 3 K below the the temperature of the superbubble in NGC 4631. When we take the 70 ${\mu}$m data point as an upper limit, the temperature decreases slightly. Therefore, the temperature difference of superbubble for NGC 4631 and NGC 891 cannot be alleviated by different SED fitting processes.⁵⁵5 For the SED modeling, we used 5 bands, 70 ${\mu}$m, 160 ${\mu}$m, 250 ${\mu}$m, 350 ${\mu}$m, and 500 ${\mu}$m while Paper I adopted 100 ${\mu}$m in addition to the 5 bands. Also, in Paper I, the 70 ${\mu}$m data point was considered as an upper limit. The higher temperature of the SED fit for NGC 4631 is driven by the data point at 100 ${\mu}$m. When we include the 100 ${\mu}$m data point from the HEDGES data, the temperature of all components of NGC 891 increases by $\sim 1$ K.

Dust in galaxy halos can be heated by the evolved stellar population or processes associated with the stellar feedback. Using B and V band photometry, Ann et al. (2011) measured a scale height for the thick stellar disk of NGC 4631 of $1.33\pm 0.25$ kpc and $1.25\pm 0.28$ kpc, and Morrison et al. (1997) report an R band scale height of $1.5-2.5$ kpc for NGC 891. It therefore seems unlikely that the stellar halo of NGC 891 is any less effective at heating dust than that in NGC 4631. The SFR surface density of NGC 4631 is only slightly higher than that of NGC 891, so an attempt to explain the difference in superbubble temperatures would need to appeal to the lower dust mass and smaller size of the superbubble in NGC 4631. It would be interesting to explore whether these properties of the NGC 4631 superbubble reflect a younger age or the more ordered magnetic field of the halo. The increased temperature of the superbubble might reflect the larger pressure required to push these field lines out of the superbubble’s path.