Hodges-Kluck 2016 ApJ 833 58

The Astrophysical Journal, 833:58 (29pp), 2016 December 10 doi:10.
3847/1538-4357/833/1/58
© 2016. The American Astronomical Society. All rights reserved.
ULTRAVIOLET HALOS AROUND SPIRAL GALAXIES. I. MORPHOLOGY

Edmund Hodges-Kluck, Julian Cafmeyer, and Joel N. Bregman
Department of Astronomy, University of Michigan, Ann Arbor, MI 48109, USA; hodgeskl@umich.edu
Received 2016 August 24; revised 2016 October 4; accepted 2016 October 5; published 2016 December 7
ABSTRACT
We examine ultraviolet halos around a sample of highly inclined galaxies within 25 Mpc to measure their
morphology and luminosity. Despite contamination from galactic light scattered into the wings of the point-spread
function, we find that ultraviolet (UV) halos occur around each galaxy in our sample. Around most galaxies the
halos form a thick, diffuse disk-like structure, but starburst galaxies with galactic superwinds have qualitatively
different halos that are more extensive and have filamentary structure. The spatial coincidence of the UV halos
above star-forming regions, the lack of consistent association with outflows or extraplanar ionized gas, and the
strong correlation between the halo and galaxy UV luminosity suggest that the UV light is an extragalactic
reflection nebula. UV halos may thus represent 106–107 Me of dust within 2–10 kpc of the disk, whose properties
may change with height in starburst galaxies.
Key words: dust, extinction – galaxies: halos – ultraviolet: galaxies
Supporting material: machine-readable tables
1. INTRODUCTION around highly inclined galaxies, but Sandin (2015) pointed out
that the light scattered into the wings of the point-spread
Dust plays an important role in galaxies as a catalyst for
function (PSF) can masquerade as a physical halo. Shinn &
interstellar chemistry, a repository of metals, and a heat source
Seon (2015) modeled several of the galaxies in HKB14
for interstellar gas. The discovery that there is about as much
including this effect, and found that some of the UV halos we
dust outside of galaxies as within (Ménard et al. 2010) suggests had reported are astrophysical while others are artificial. In
that dust also plays an important role in the disk–halo cycle principle, eRN are visible at other wavelengths, and some of
(however, see Smith et al. 2016, who argue for a smaller the Hα attributed to extraplanar diffuse ionized gas (eDIG) may
amount of halo dust). Dust is primarily formed in the actually be reflected light (Ferrara et al. 1996; Seon et al. 2014).
atmospheres of evolved stars or supernova remnants, so However, UV halos are especially promising because the
extragalactic dust was likely transported out of the disk by scattering cross-section is high in the UV for typical dust
gaseous flows (e.g., Ferrara et al. 1991). Indeed, dusty outflows compositions and the sky is relatively dark.
have been detected around edge-on galaxies, including In HKB14 we examined galaxies with both Swift UVOT
filaments seen in extinction (Howk & Savage 1997, 1999) (Roming et al. 2005) and GALEX (Martin et al. 2005) observa-
and dust emission at several kpc above the midplane tions, regardless of physical properties. In a series of three papers
(McCormick et al. 2013; Meléndez et al. 2015). Still, it we now take a more systematic approach to measuring the
remains unclear how and when the dust gets into the halo, how physical properties of the UV halos around highly inclined spiral
long it remains there, and whether it is altered in the process. galaxies. In this paper (Paper I) we focus on galaxies within about
These issues motivate further studies. 25 Mpc to determine the frequency of physical UV halos, their
Extragalactic dust can be studied through extinction, morphology, and whether they are eRN. The subsequent papers
emission, or reflection (scattering). In the first case, extraplanar will present a catalog of total halo fluxes for a wider sample
dust lanes can be seen against the thick disk of stellar light where morphology cannot necessarily be measured (Paper II),
(Howk & Savage 1997), but above a few kpc from the galaxy and models of the spectral energy distributions (SEDs) for those
midplane the extinction must be measured toward background galaxies with the highest quality data (Paper III).
continuum sources, such as quasars. Since there are few The remainder of this paper is organized as follows.
quasars behind any one galaxy, Ménard et al. (2010) stacked Section 2 describes our sample and data sources, and Section 3
many sources to measure the extinction to beyond 1 Mpc. In is focused on correcting for galactic light scattered into the PSF
contrast, thermal emission or reflection will only be seen near wings. In Section 4 we describe the morphology of the UV
the galaxy, since both processes need a nearby light source. halos and compare them to diffuse halos at other wavelengths,
This makes edge-on systems the ideal laboratories to isolate and in Section 5 we use information from the halo fluxes in
extraplanar dust emission and scattered light. each filter to characterize the UV halos. We interpret the results
The possibility of extragalactic reflection nebulae (eRN) in Section 6 to argue that the UV halos are likely eRN and
around edge-on galaxies was raised by Ferrara et al. (1996), connect the UV measurements to extinction and emission
and Hoopes et al. (2005) reported the discovery of diffuse measurements. Section 7 summarizes the paper and presents
ultraviolet (UV) light consistent with an eRN above the disks our main conclusions.
of the edge-on starburst galaxies M82 and NGC253. This
conclusion was reinforced by Hutton et al. (2014), and Seon
2. DATA
et al. (2014) detected a UV halo around NGC891 that they
argue is also an eRN. In Hodges-Kluck & Bregman (2014, Our goal is to measure the structure of the diffuse extraplanar
hereafter, HKB14) we reported that UV halos are ubiquitous UV light, so we limit the sample to nearby, highly inclined,
1
The Astrophysical Journal, 833:58 (29pp), 2016 December 10 Hodges-Kluck, Cafmeyer, & Bregman
spiral galaxies. The initial sample (drawn from the HyperLeda PSF-wing contamination in the measured halo fluxes. We want
database1; Makarov et al. 2014) includes galaxies within to obtain accurate 2D maps of the halo rather than only a profile
25 Mpc with an inclination angle of i 75°, a Hubble type of along the minor axis, so the typical approach of convolving a
Sa or later, and an absolute B-band magnitude of model with the PSF is not sufficient.
MB<−20 mag. The distance limit is based on the need for
high signal-to-noise ratio (S/N) in small regions. We excluded
some galaxies for which the inclination angles or types were 3.1. Measured PSFs
obviously incorrect, and for our distance cut we used the The GALEX and Swift calibration teams have provided PSFs
redshift-independent distances from NASA/IPAC Extragalac- measured to radii of about 60 and 30 arcsec, respectively3, but
tic Database2 instead of the HyperLeda values. The initial for the galaxies in this work the PSFs need to extend to
sample includes 78 galaxies. 200–900 arcsec. Extrapolating the wings from the publicly
Each galaxy must have moderately deep data in at least one available PSFs may not be sufficient, so we measured PSFs
of the GALEX or Swift UVOT filters: GALEX FUV (λ1516 Å) from bright point sources to better characterize the wings.
and NUV (λ2267 Å), or the UVOT uvw1 (λ2600 Å), uvm2 For GALEX, we used deep observations of the quasars
(λ2246 Å), and uvw2 (λ1928 Å) filters. Most of the data come 3C273 (30 ks, tile GI4_012003_3C273) and PKS2155−304
from the GALEX or Swift archives, but we also obtained new (32 ks, tile PKS2155m304), which are relatively isolated on the
UVOT data for several targets through the Swift program sky and not bright enough to produce strong ghost images or
GO1013198. Based on our prior work, we expect a UV halo diffraction spikes. These quasars are too bright to measure
flux that is about 1% of the total galaxy flux, so to accumulate extended PSFs with Swift, so we used a combined image of
enough S/N we require exposure times greater than 2–4 ks in Mrk501, which has an effective exposure of 129 ks in uvw1,
either GALEX filter or 5–6 ks in any UVOT filter (this excludes 131 ks in uvm2, and 170 ks in uvw2. The data were processed
the 600 s exposures from the GALEX all-sky imaging survey). as described above and in HKB14, except that we used at least
We excluded galaxies that are in regions with strongly variable 15 sources to compute the astrometric solution for each image
Galactic cirrus or are in very crowded fields. Swift targets must and we rejected images where the source was outside a few
also fit within the 17×17 arcmin field of view for accurate arcminutes from the chip center. The observation IDs we used
background subtraction. The 26 galaxies that meet this criteria are given in the Appendix.
form our working sample, and the basic properties of these We then measured the PSF by constructing an azimuthally
objects are given in Table 1. The UV exposure times and 3σ symmetric radial profile to r>100 arcsec. The actual PSF is
point-source sensitivities are given in Table 2. The observation not azimuthally symmetric, but simulations show that unless it
IDs used in this paper are given in the Appendix. is very asymmetric (to an extent that would be noticeable on
Our reliance on archival data means that the sensitivity and the image), this approximation does not noticeably change the
filter coverage vary widely among the galaxies. We also note measured PSF-wing contamination for any galaxy with a
that a disproportionate number of galaxies are in the Virgo projected size of about an arcminute or more (even at smaller
Cluster, which has been observed extensively. Some of the sizes, it will not usually make a difference). The background
galaxies are included in published work on UV halos (Hoopes level is crucial to measuring the small amount of light scattered
et al. 2005; Hodges-Kluck & Bregman 2014; Hutton by the PSF to large radii, so we masked all point sources
et al. 2014; Seon et al. 2014; Shinn & Seon 2015; Baes & detected using the Swift task uvotdetect with a 3σ detection
Viaene 2016). threshold (based on SExtractor; Bertin & Arnouts 1996) other
We processed the data for each galaxy as described than the target. Since these sources are much fainter than the
in HKB14, including cleaning the UVOT and GALEX images target, we assume that masking to the nominal 99% encircled
of diffuse artifacts in each exposure. We combined the Swift energy radius is sufficient.
exposures for each filter into a single, exposure-corrected The measured profiles are shown in Figure 1, along with
image. We masked all the point sources outside the galaxy that fitted profiles of the form Ar− x+B, where A, B, and x are
were detected at 3σ or greater, and we also masked each source determined using least-squares fits to the data between the
detected at 4σ or greater in any one filter in all the remaining dashed vertical lines in Figure 1 (in each plot, B has been
filters. This helps to exclude stars that are close to the galaxy subtracted). For comparison, we also plot a fitted profile where
but are not formally detected at different wavelengths or in x≡2. The extrapolation based on the best-fit model begins at
shorter observations. the midpoint between the dashed vertical lines, but our results
are insensitive to where we match the extrapolated profile to the
3. EXTENDED PSF WINGS AND HALO LIGHT measured one. The FUV is close to x=2 at all radii, but the
other filters concentrate light within about 30 arcsec, with the
Sandin (2015) showed that the diffuse light around edge-on subsequent decay being steeper than x=2.
galaxies consists in part of light from the galaxy image The best-fit exponents are given in Table 3. The fit depends
scattered into the wings of the PSF (the Airy pattern). To strongly on B, so we verified that the background level in the
measure and subtract this component, a PSF must be measured fitting zone was consistent with the background measured
to at least twice the angular size of the galaxy. Hereafter, we outside the fitting zone. We also repeated the analysis using a
refer to this light as “PSF-wing contamination.” Shinn & Seon 2σ threshold for point source removal and measured x values
(2015) were the first to address this issue in the GALEX FUV that are consistent with the earlier measurements. Finally, we
filter, and here we extend their analysis by measuring and verified that the PSFs successfully clean the outer regions in the
extrapolating the PSFs for each filter in order to determine the source images and other images of bright point sources (within
1 3
http://leda.univ-lyon1.fr/ See http://www.galex.caltech.edu/researcher/techdoc-ch5.htmland http://
2
http://ned.ipac.caltech.edu/ heasarc.gsfc.nasa.gov/docs/heasarc/caldb/swift/docs/uvot/.
2
Table 1
Basic Galaxy Parameters
Name Type T i MB d vrot E (B - V ) B−V mK M* LHα (1040 SFR(IR)

(deg) (mag) (Mpc) (km s−1) (mag) (mag) (mag) (10 Me)
10
erg s−1) (Me yr−1)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
Starbursts
NGC 0253 SABc 5.1 76 −21.23 3.25 189.8 0.019 0.69 3.772 4.36 9.59 3.97
M82 Scd 7.5 76.9 −20.13 3.93 65.6 0.138 0.68 4.665 1.83 15.0 9.42
NGC 4631 SBcd 6.6 85 −22.42 6.02 138.4 0.015 0.39 6.465 0.98 15.0 1.03
NGC 3628 Sb 3.1 79.3 −21.54 10.89 215.4 0.024 0.68 6.074 4.47 4.59 3.30
NGC 4666 SABc 5.1 69.6 −21.10 17.28 192.9 0.022 0.64 7.055 3.10 16.8 4.83
NGC 3079 SBcd 6.6 84 −21.56 19.28 208.4 0.01 0.53 7.262 2.91 16.9 8.17
NGC 5775 SBc 5.2 83.2 −21.09 20.34 187.2 0.037 0.66 7.763 2.45 0.014 3.97
NGC 4388 Sb 2.7 85 −22.13 20.5 171.2 0.029 0.57 8.004 1.55 2.35 3.16
Normal Spirals
NGC 0055 SBm 8.8 80 −20.09 1.94 58.7 0.012 0.33 6.249 0.09 3.42 0.06
NGC 0891 Sb 3.1 89.5 −20.37 9.96 212.1 0.058 0.70 5.938 3.99 5.30 2.43
NGC 2683 Sb 3.0 82.8 −20.42 10.08 202.6 0.029 0.75 6.328 2.98 5.56 0.41
NGC 4517 Sc 6.0 86 −21.46 10.56 139.6 0.021 0.53 7.329 0.73 L 0.34
NGC 4565 Sb 3.3 87 −22.55 12.18 243.6 0.014 0.68 6.060 5.65 2.20 0.80
NGC 4096 SABc 5.3 80.5 −20.39 12.68 144.8 0.016 0.50 7.806 0.088 5.18 0.56
NGC 4313 Sab 2.1 90 −20.16 14.62 117.6 0.033 L 8.468 L L 0.37
NGC 3623 Sa 1.0 76 −21.02 12.77 231.2 0.022 0.78 6.066 7.17 5.76 0.38
NGC 5907 SABc 5.3 88 −21.08 16.37 226.6 0.009 0.62 6.757 5.03 14.1 2.04
NGC 4216 SABb 3.0 85 −20.80 16.78 244 0.028 0.83 6.524 6.06 L 0.44
NGC 4607 SBbc 4.0 83 −20.18 17.78 98.9 0.028 0.75 9.584 0.60 1.17 0.68
NGC 4522 SBc 5.9 79.2 −20.91 18.29 96.4 0.018 L 9.8 L 1.67 0.40
NGC 0134 SABb 4.0 76.8 −21.49 18.71 220.2 0.016 0.69 6.844 5.95 L 4.51
NGC 4157 SABb 3.3 90 −19.88 18.7 188.9 0.019 0.64 7.363 3.03 8.11 2.71
ESO358-063 Scd 6.9 75.6 −20.34 18.98 135 0.005 0.7 9.144 0.61 L 0.87
NGC 4217 Sb 3.1 81 −20.08 19.37 187.6 0.015 0.75 7.582 3.67 3.08 L
NGC 4330 Sc 6.0 78.9 −20.02 19.61 115.7 0.021 L 9.51 L L 0.36
NGC 3044 SBc 5.5 84 −20.32 22.48 152.6 0.022 0.6 8.982 1.01 15.1 2.77
NGC 5170 Sc 4.9 87 −21.13 26.8 244.7 0.07 0.7 7.628 7.46 L 0.79
Note. Columns (1) Name, (2) Morphological type, (3) Morphological type code, (4) Inclination angle, (5) B magnitude, (6) Distance, (7) Circular rotational velocity,
(8) Foreground Galactic extinction, (9) B−V color, (10) K-band magnitude from the 2MASS Extended Source Catalog Schlafly & Finkbeiner (2011), (11) Stellar
mass using relation from Bell & de Jong (2001), (12) Hα luminosity, (13) Star formation rate estimated from the Kennicutt (1998) relation SFR=4.5×10−44 LIR
Me yr−1. LIR was measured as defined by Rice et al. (1988) LIR = 5.67 ´ 10 5dMpc 2
(13.48f12 + 5.16f25 + 2.58f60 + f100 ) L  , where the fluxes at 12, 25, 60, and
100 μm are in Jy from the IRAS catalog.
References. Values from the NASA/IPAC Extragalactic Database, (http://ned.ipac.caltech.edu/), HyperLeda Makarov et al. (2014), and the IRAS catalog
(http://irsa.ipac.caltech.edu/Missions/iras.html).
∼10 arcsec an azimuthally symmetric PSF is a poor match to in the best-fit x between combined images taken in different
the data). months. We found no clear variability, but some low level of
There are two caveats to this analysis. First, the PSF variability beyond 50 arcsec is possible. We note that the Swift
becomes increasingly distorted with increasing distance from PSF core does vary with time in the sense that the full-width at
the optical axis, so for images “far” from the optical axis the half maximum depends on the temperature of the UVOT
assumption of radial symmetry is badly wrong (where the focusing rods.4 However, we did not find any variation in the
critical distance depends on the instrument). The PSF is best-fit exponent for the PSF wings.
naturally hard to measure to large radii far from the optical axis, Our results are consistent with the best-fit exponents found
and there are few bright sources that are both near the chip edge in other sources where the PSF cannot be measured to such
and have very deep GALEX or Swift data. Thus, for galaxies large distances, so we believe that the fits are reasonably
near the chip edge the measured PSFs may not accurately accurate.
account for the PSF-wing contamination.
Second, the PSF may vary with time (Sandin 2015, although
3.2. Estimating the PSF-wing Contamination
the variability is probably much larger for ground-based
instruments). We could not make useful tests with GALEX We estimated the PSF-wing contamination for each image
data because the bright sources we used for comparison by creating a model using the input image and the extrapolated
typically had different PSF cores that were obviously due to the PSF models, using the best-fit exponent for each filter (Table 3).
different positions on the chip. For Swift we selected all We first clipped the input image to the region that just contains
exposures of Mrk501 in a given filter where the source was
4
within 1 arcmin of the chip center, and searched for variability https://heasarc.gsfc.nasa.gov/ftools/caldb/help/uvotapercorr.html
3
Table 2
Observations
Name Exposure Time (s) Sensitivity (AB mag) Hα Radio 21 cm

FUV NUV uvw2 uvm2 uvw1 FUV NUV uvw2 uvm2 uvw1
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)
Starbursts
NGC 0253 3281 15043 L L L 25.0 26.1 L L L L 5 13

M82 14707 29422 91603 147713 92788 25.9 26.4 26.8 25.7 26.5 10 21 24
NGC 4631 3147 3147 1377 5474 1123 24.9 25.1 24.0 24.7 23.6 10 21 19
NGC 3628 5812 17076 8095 5530 3546 25.4 26.0 25.2 24.8 24.3 L 21 6
NGC 4666 5940 5940 19561 19610 9858 25.4 25.3 24.7 25.7 25.1 L 21 18
NGC 3079 16108 16108 8512 7252 273 26.3 26.1 25.4 25.1 22.5 23 21 7
NGC 5775 2587 5770 14978 23677 4180 24.6 25.4 25.6 25.8 24.5 4 21 8
NGC 4388 2538 4993 10331 7334 5331 24.6 25.3 25.4 25.0 24.7 16 21 3
Normal Spirals
NGC 0055 29347 30691 L L L 25.5 26.5 L L L L L 20

NGC 0891 6047 6283 15353 15200 9787 25.0 25.2 25.7 25.5 25.2 16 21 14
NGC 2683 1600 1600 6407 5644 4824 24.3 24.5 25.0 24.8 25.3 16 21 17
NGC 4517 1906 6403 L L L 23.8 25.5 L L L L L
NGC 4565 12050 L L L L L 25.9 L L L L 21 25
NGC 4096 1650 1650 2670 3575 L 24.3 24.7 24.6 24.5 L 10 L L
NGC 4313 3862 3862 L L L 24.2 25.2 L L L L L L
NGC 3623 1656 1656 5300 7122 3193 24.3 24.6 25.2 25.1 24.5 10 5 6
NGC 5907 1543 5423 31888 45601 26482 24.2 25.5 26.3 26.3 26.0 15 21 1
NGC 4216 1672 2604 L L L 24.3 24.9 L L L 16 5 3
NGC 4607 2701 6181 L L L 24.7 25.5 L L L 11 L 3
NGC 4522 2496 2496 L L L 24.7 24.9 L L L 11 L 9
NGC 0134 5998 6336 4032 L L 25.1 24.9 24.6 L L L L L
NGC 4157 2494 1288 L L L 24.7 24.5 L L L 23 21 22
ESO358-063 2456 3197 L L L 24.7 25.2 L L L L L L
NGC 4217 11774 11774 L L L 25.8 25.9 L L L 15 21 1
NGC 4330 3862 3862 2429 2429 2279 25.1 25.2 24.3 24.1 24.0 L L 3
NGC 3044 1676 1676 2910 2910 2585 24.3 24.6 L L L 4 21 12
NGC 5170 1606 3307 L L L 24.0 24.9 L L L L L 2
Note. Columns (2–6) Exposure times, (7–11) 3σ point-source sensitivity, (12–14) References for non-UV data used here.
References. Hα and radio references: (1) Allaert et al. (2015), (2) Bottema et al. (1987), (3) Chung et al. (2009), (4) Collins et al. (2000), (5) Condon (1987), (6)
Haynes et al. (1979), (7) Irwin et al. (1987), (8) Irwin (1994), (9) Kenney & Koopmann (2001), (10) Kennicutt et al. (2008), (11) Koopmann et al. (2001), (12) Lee &
Irwin (1997), (13) Lucero et al. (2015), (14) Oosterloo et al. (2007), (15) Rand (1996), (16) Rossa & Dettmar (2003a), (17) Vollmer et al. (2016), (18) Walter et al.
(2004), (19) Weliachew et al. (1978), (20) Westmeier et al. (2013), (21) Wiegert et al. (2015), (22) Yim et al. (2014), (23) Young et al. (1996), (24) Yun et al. (1994),
(25) Zschaechner et al. (2012).
the galaxy. We did not use an objectively defined region clumps. The combination of these issues leads to an over-
because of the varied disk morphologies and inclinations in our estimate of the PSF-wing contamination by about 10%
sample, but the galaxy regions are conservative in that they immediately adjacent to the galaxy, and which quickly declines
extend to at least 2 kpc above the midplane. We then convolved with height. We also investigated the error due to using a
the clipped image with a PSF model that was extrapolated to at symmetric PSF in the FUV, where the PSF is measured to over
least twice the angular diameter of the galaxy. 100 arcsec. We convolved model images with angular sizes
The resulting image shows the amplitude of the PSF-wing smaller than this with our model PSFs and the measured PSF.
light at each position around the galaxy, but the total flux in the This leads to differences of 0%–3% in the PSF-wing light
model image is too low and the light is too smeared out because measured around the galaxy. On the scale of the whole halo,
we do not deconvolve the galaxy image before convolving it these effects are very modest.
with the extrapolated PSF (the galaxy is too complex for The model image can then be used in conjunction with the
accurate deconvolution). The former effect is corrected by original image to measure the PSF-wing contamination at any
multiplying the fake image by the ratio of the flux in the galaxy point in the halo, and we also subtract the model from the
in the original and fake images (this is not an exact correction, original to create corrected halo images. The PSF-wing
but simulations of exponential disk models show that it contamination differs for each galaxy and filter, but it ranges
underestimates the PSF-wing flux by much less than the from 1% to 90% of the total extraplanar flux in our sample. The
statistical uncertainties for our sample). The latter effect cannot contamination is generally larger in the FUV than the other
be easily corrected, but we find (again using simulations of filters. The importance of the correction is illustrated in
exponential disk models) that the most severe impact is within Figure 2 for NGC3079, where we show FUV images before
about 10 pixels of bright regions in each galaxy, and in practice and after subtracting the model image. We verified the Sandin
the galaxy regions we define extend at least this far from bright (2015) proposition that convolving the galaxy image with a
4
Figure 1. Measured, radially averaged PSF profiles for each filter (filled circles) and the best-fit profile (black solid line). The fit is based on points between the vertical
dashed lines, and the profile is extrapolated from the data starting halfway between these lines. The statistical 1σ error bars on the fit are shown as dashed lines. For
comparison, the r−2 profile is shown as a red line. The last panel shows the extrapolated profiles to a radius of 300 arcsec for each filter.
Table 3
PSF Wing Exponents
Filter Exponent Filter Exponent

FUV 2.05±0.01 uvm2 2.4±0.1
NUV 3.1±0.3 uvw1 2.55±0.08
uvw2 2.4±0.1 L L
Note. Best-fit exponents (x in r− x) and statistical errors from the PSF profile
fitting.
Figure 2. Left: FUV image of NGC3079 before correcting for galactic light
PSF extrapolated to more than twice the angular size of the scattered into the PSF wings. The contours are 3, 4, and 5σ above background.
galaxy does not significantly alter the PSF-wing contamination. Right: the same image after correction for PSF-wing contamination. The galaxy
image used to compute the correction has been masked.
In Section 5 we assess the reliability of our PSF-wing
subtraction.
4.1. Existence of UV Halos
We use UV images where the PSF-wing contamination has
4. EXTRAPLANAR ULTRAVIOLET MORPHOLOGY been subtracted. Figures 3 and 4 show the images in each filter
In this section we describe the qualitative structure of the for the normal and starburst galaxies, respectively. In these
diffuse extraplanar UV light in our sample and compare it to images, the galaxy region used to estimate the PSF-wing
the diffuse structure seen at other wavelengths where the source contamination is masked, so we superimpose the unsmoothed
of the light is basically understood. The initial work on UV image of the galaxy for that filter in this region (in blue).
halos (Hoopes et al. 2005) focused on galaxies with galactic Outside of this region, point sources have been masked, the
superwinds, and in this section we divide our sample into image has been smoothed with a Gaussian kernel (with σ = 3–8
galaxies with known winds (starburst galaxies) and those pixels, depending on the angular size of the galaxy), and the
without (normal galaxies). image has been clipped to 1σ above the background. The
5
Figure 3. UV halos (orange) around normal galaxies in the GALEX and UVOT filters, shown with increasing wavelength from left to right. The final column shows
the combined data, referenced to the GALEX resolution. The orange maps have been corrected for the PSF-wing contamination from the galaxy (image shown in blue)
and point sources have been masked. For each filter, the contours are the 2, 4, 8, and 16σ contours above the background. A 2 arcmin line is shown at the bottom left
of each row, and the 3σ point-source sensitivity for the map is shown at the bottom right of each panel. The composite image only shows the 3σ contour.
6
Figure 3. (Continued.)
7
contours show the 2, 4, 8, and 16σ level above the background. We investigated the contribution of stellar clusters or
For each image we also list the 3σ point-source sensitivity in background sources to the diffuse light in those galaxies with
AB magnitudes (Oke & Gunn 1983), which shows the Hubble Space Telescope (HST) observations in the U band or
differences in depth between filters. bluer. For these objects, we identified point-like sources in the
Although the UVOT and GALEX images have different HST maps that were not already identified and removed in our
PSFs, artifacts, and systematic uncertainties, the morphology of maps and predicted a Swift or GALEX flux for each, assuming
the residual extraplanar UV light is similar between filters for that each source has the spectrum of a B star. We then
most galaxies. Thus, in the final columns of Figures 3 and 4 we compared the flux to the total diffuse UV flux, and found that it
show a composite image for each galaxy made by summing can explain at most a few percent of the light. The exception
each filter (referenced to the effective area of the uvm2 filter). that proves the rule is NGC4522, where the Virgo cluster is
These images have roughly twice the S/N of the images in stripping the ISM (Kenney & Koopmann 1999) and there are
each filter, so they highlight the extent of the halo. Remarkably, known stellar clusters in the halo. In this case, most of the flux
every galaxy in the sample has a UV halo extending at least to in the FUV comes from these clusters. Even NGC4522 has a
6 kpc above the midplane. Considering that our selection diffuse component, however, so we conclude that all of these
criteria were distance, luminosity, and inclination, this suggests galaxies have true diffuse UV halos.
that UV halos are a common feature of luminous spiral
galaxies. 4.2. Ultraviolet Morphology
There are some differences between filters for several There are three qualitative morphological metrics we use to
galaxies that may be artificial. For example, in the uvw2 compare the UV halos in our sample: filamentary structure,
images of NGC4666 and NGC5775 (Figure 4), the halo concentration of the UV light (both in the vertical direction and
appears to be more prominent than in the uvm2 or FUV images, along the disk), and symmetry across the midplane. In general,
even though the uvm2 has a higher sensitivity. This may these are consistent among filters for a given galaxy.
indicate residual light in the central 2 arcmin halo ring in the Filaments—filaments are most easily visible in the FUV, and
large-scale uvw2 scattered-light artifact (HKB14), which could they are seen in most starburst galaxies (M82, NGC 253,
mimic a halo in an image centered on the galaxy. The smaller NGC 3079, NGC 3628, NGC 4666, and NGC 5775), as well as
field of view of the UVOT can also lead to differences: in the two normal galaxies (NGC 4522, NGC 4330). The filaments in
relatively shallow UVOT images of NGC4631 (Figure 4), the the latter two are associated with gas stripped by ram pressure
halo is both less prominent and more symmetric than in the in the Virgo cluster. We cannot rule out the presence of
GALEX images. This probably results from the “background” filaments in non-stripped normal galaxies, but those galaxies
regions in the UVOT being inside the UV halo itself, a scenario with high S/N images do not have filaments.
that is supported by the GALEX images. In other cases, Concentration—the UV halos differ in extent, both in
differences between the morphology in each filter likely galactocentric radius and projected height. For example, in
indicate physical differences. For example, in M82 and NGC891, NGC2683, and NGC3044 most of the light is
NGC253 (Figure 4), the NUV halo appears more like a thick concentrated near the center of the galaxy (Figure 3), whereas
disk than the FUV halo, which is more filamentary in nature. in NGC4216 or NGC5907 the surface brightness is compar-
This morphology cannot be explained by PSF-wing contam- able across most of the disk. We examined the degree of
ination or filter artifacts. concentration by dividing the UV halo at the midplane and
8
Figure 4. UV halos (orange) around starburst galaxies. The format is the same as in Figure 3.
9
Figure 5. Combined images from each filter stacked in the first quadrant and shown relative to the optical radius, R25. The contour represents the combined 3σ
threshold above background, and the dimensions of each image are 2 R25×2R25. The dashed lines show R25. Starburst galaxies are labeled with red names. The final
two panels show the combined normal and starburst galaxies, respectively, with scale heights measured as a function of galactocentric radius shown as points. The
“normal” stack excludes NGC4330 and NGC4522, which are being stripped by ram pressure. The dip at the center of NGC4631 is due to masking a satellite galaxy,
but it is accounted for in the starburst stack and does not cause the central dip in the scale height there.
projected galactic center into four quadrants, which were outskirts. The radial or vertical concentration is also unrelated
summed. Figure 5 shows the resulting images. Each panel has (within each subsample) to the total halo luminosity. The radial
dimensions of 2 R25×2 R25, where we use the HyperLeda concentration also appears to be unrelated to the vertical
values for R25. The final two panels show the sum of all of the concentration, and the normal galaxies appear to span a narrow
normal and starburst galaxies, respectively, where the images range in vertical extent below R25.
were reprojected and stacked on a common axis relative to R25. Bilateral symmetry—we determined the degree of bilateral
We excluded the stripped galaxies NGC4330 and NGC4522 symmetry by folding images across the midplane. Most
from this stack. galaxies show some asymmetry, although in most cases the
There are two aspects to the concentration: the extent of the visible structure is symmetric while the flux is not, which could
light (as measured by the 3σ contour in Figure 5) and its be a function of the inclination angle. Among the normal
distribution. Among the UV halos of normal galaxies, 6/18 galaxies the bilateral asymmetry is only pronounced for
extend in the radial direction to R25 and only one (NGC 4522) NGC134, NGC4330, and NGC4522, which are experiencing
extends beyond 0.5 R25 in the vertical direction. In contrast, all ram-pressure stripping (Kenney & Koopmann 1999; Abramson
of the starburst halos (8/8) extend to R25 in the radial direction, et al. 2011). Among the starburst galaxies there is asymmetry in
and 6/8 extend to 0.5 R25 in the vertical direction. In terms of the filamentary structures, but overall the halos are nearly
distribution, for most normal galaxies the UV light is brightest symmetric. An exception is NGC4631, in which the northern
and most vertically extended above the galactic center, with a part of the halo is somewhat more extended. NGC3628 is not
modest decline with increasing galactocentric radius. This is bilaterally symmetric, but its halo appears to be warped; if we
consistent with a disk-like structure. The starburst galaxies invert the image along the axis of the midplane, it becomes
other than NGC4631 are also less concentrated and filamen- nearly symmetric. A hint of a similar warp can be seen in the
tary structure at larger heights is superimposed. The similarity combined image of NGC4388.
in sensitivity and distance between several of the starburst and
normal galaxies (e.g., NGC 3079 and NGC 5907) indicates that 4.3. FUV–NUV Color
UV halos are intrinsically brighter and more morphologically In addition to structures visible in each band, we looked for
complex around starburst galaxies. spectral structure in the UV halos. In regions where the halo is
When compared to near-IR images, the radial extent of the detected in all five filters we can construct SEDs, but most
UV halo along the major axis (relative to R25) is unrelated to galaxies in the sample do not have enough data. Thus, we adopt
the prominence of the bulge, so it is unlikely that the radially the FUV–NUV color as a proxy. Figures 6 and 7 show the
concentrated UV halos are merely light from the bulge FUV–NUV color in regions where the flux is detected at 2σ
10
Figure 7. The FUV–NUV color for starburst galaxies. The plot style is the
same as in Figure 6.
structures within the wind that are visible in both the FUV
and NUV images do not appear in the FUV–NUV color map.
In other words, small UV structures in the wind appear to have
the same color as the rest of it. It is also of interest that the bluer
Figure 6. The FUV–NUV color (per square arcsec) for normal galaxies. The regions in the halo of NGC4631 are asymmetric across the
FUV and NUV images used to create these maps were smoothed and clipped at midplane, although both sides are blue in the central region
the mutual 2σ level (i.e., where emission is detected at 2σ above background in where the wind occurs.
each image). The GALEX pixel size is 1.5 arcsec.
4.4. Multiwavelength Comparison
above background in each filter for normal and starburst
We compare the diffuse UV halos to diffuse extraplanar light
galaxies, respectively. In these images, the smoothing kernel is
in other bands, as well as the disk emission in the same bands.
usually larger than in Figures 3 and 4. The colorbar in the plots
These include Hα, soft X-rays, 160 μm emission, radio
is such that blue regions correspond to a blue spectrum (small continuum (at 1.4 or 4.8 GHz), and 21 cm emission from
FUV–NUV) and red regions correspond to a red spectrum neutral hydrogen. The Hα traces the diffuse ionized gas in and
(large FUV–NUV). However, values at the edges are not above the disk, and the presence of extraplanar diffuse ionized
reliable as they are sensitive to the background subtraction gas (eDIG) indicates disk porosity and ongoing star formation
(e.g., the red boundaries on the halo of NGC 55 in Figure 3). (it may also produce UV emission nebulae through ionized
There is no obvious structure in the FUV–NUV color maps helium). We use Hα images from several atlases (Lehnert &
for most normal galaxies. There appears to be a tendency for Heckman 1996; Rand 1996; Collins et al. 2000; Rossa &
the Sc-type galaxies (NGC 5907, NGC 4522, NGC 4330, and Dettmar 2003b). Diffuse soft X-rays trace hot gas, and X-ray
NGC 3044) to have bluer halos than the rest (which are Sb, halos seen near the disk are primarily attributed to outflows
except for NGC 55), but the strength of this connection cannot connected to star formation (Li & Wang 2013). Within the
be assessed with the current sample. This is especially true disk, some of the X-rays also come from X-ray binaries. We
since NGC4522 and NGC4330 are being stripped of their gas use X-ray data from the Chandra and XMM-Newton archives,
by ram pressure from the Virgo intracluster medium. For and the presence of X-ray halos is determined from the
galaxies where the S/N is reasonably high across the disk literature (Strickland et al. 2004; Tüllmann et al. 2006; Li &
(NGC 55, NGC 891, NGC 4217, and NGC 4330), the color Wang 2013). The 160 μm images trace thermal emission from
does not change much across with projected galactocentric dust, and we obtain them from the Herschel archive. We use
radius. However, in the starburst galaxies the filaments and 160 μm because it is near the expected peak of the modified
regions associated with galactic winds are bluer than the rest of blackbody curve for thermal emission, and the spatial
the halo. The underlying FUV–NUV color differs between resolution at longer wavelengths is considerably worse. The
galaxies, with NGC3628 and NGC4388 having generally 160 μm images are dominated by the disk and the PSF is about
redder halos; both of these galaxies are classified as Sb type, 6 arcsec, so we adopt the same approach to clean the
whereas the others are Sc or Sd. Combined with the normal overlapping Airy patterns and isolate extraplanar emission
galaxies, this suggests that bluer galaxies tend to have bluer (except in NGC 4522). We use the PSF from Aniano et al.
halos, but the sample size is small. In M82 the FUV–NUV (2011), and in some cases the galaxy box is smaller than for the
color tends to decrease with height, but the filamentary UV. We also refer to McCormick et al. (2013) regarding
11
extraplanar polycyclic aromatic hydrocarbon (PAH) emission not always as extensive. NGC55, NGC891, and NGC4096
and Howk & Savage (1997, 1999) for the presence of have the most convincing 160 μm halos, but their prominence
extraplanar dust seen in absorption as filaments in the optical does not appear to be connected to that in Hα or the radio
band. The radio continuum emission comes from cosmic rays continuum. Likewise, the prominence of the radio halos seems
produced in supernova remnants, which diffuse out of the disk. unrelated to the prominence of the UV halo or extraplanar
We primarily refer to data taken as part of the CHANG-ES emission in other bands. The only normal galaxies in
program (Wiegert et al. 2015), but some images come from McCormick et al. (2013) with extraplanar PAH emission are
prior atlases (especially Condon 1987). Finally, we use or refer NGC55 and NGC891. In the former case, the PAHs do not
to high resolution 21 cm maps from a variety of papers follow the UV halo, whereas in the latter they do. NGC891
(Weliachew et al. 1978; Haynes et al. 1979; Irwin et al. 1987; and NGC4217 have high latitude dust seen by Howk &
Irwin 1994; Yun et al. 1994; Lee & Irwin 1997; Kenney & Savage (1999). Although these two galaxies do have brighter
Koopmann 2001; Walter et al. 2004; Oosterloo et al. 2007; UV halos than others in the Howk & Savage (1999) sample, the
Chung et al. 2009; Zschaechner et al. 2012; Westmeier UV halos in the four galaxies without high latitude dust
et al. 2013; Yim et al. 2014; Allaert et al. 2015; Lucero absorption have about the same vertical extent and radial
et al. 2015; Vollmer et al. 2016). These maps often show concentration.
filaments of cold gas or warped disks. Starburst UV halos have filamentary structure that is also seen
Hα, soft X-rays, thermal dust emission, and the radio at other wavelengths. This is especially true in Hα, although a
continuum are associated with ongoing star formation, whereas similar pattern is seen in X-ray and radio halos. There are some
H I need not be. Figures 8 and 9 show the UV contours from disagreements in extraplanar morphology among these bands
above overlaid on Hα, X-ray, 160 μm, radio continuum, and (such as in M82, NGC 4631, or NGC 4388), but there is a clear
21 cm images for normal and starburst galaxies, respectively. In connection between the UV filaments and those seen at other
some cases there is no publicly available data, but we can wavelengths. The same is true for the extraplanar dust emission,
usually rely on images in the literature (this is especially true although not every filament is associated with 160 μm emission.
for 21 cm data). Galaxies are not included in these figures if However, unlike at other wavelengths the UV halos span the
they do not have publicly available FITS files in at least two disks of the starburst galaxies, and this non-filamentary structure
bands. As with the UV images, the S/N is variable among the is not present in Hα, X-rays, thermal dust emission, or radio
galaxies and bands, which limits the power of this comparison. continuum. In general the PAHs do not follow the UV halos
Extraplanar diffuse emission: frequency—about 40% of (McCormick et al. 2013). Some of the same filaments are visible
non-starburst galaxies have eDIG (Rossa & Dettmar 2003a), in PAHs, but the coincidence of UV and PAH filaments is worse
and a similar number of galaxies in our sample with data from than at other wavelengths, and the diffuse components above the
these bands (Table 6) have eDIG (38%), radio (45%), or soft disk have different shapes.
X-ray halos (40%). However, 67% of galaxies with 160 μm The connection between the 21 cm morphology and the UV
images have extraplanar emission (sometimes at lower heights halos is unclear. We note that the sensitivities and resolutions of
than the UV halos; Figure 8). The sample overlap with the H I images in Figure 9 differ, so the NH=5×1020 cm−2
McCormick et al. (2013) is not large enough to determine the contour (assuming optically thin hydrogen) is shown as a
fraction with extraplanar PAH emission. In contrast, UV halos shorthand for the transition from disk to halo gas. The H I and
are seen around every normal galaxy. Several of the galaxies in UV morphology are similar for the galaxies undergoing ram-
our sample were also investigated by Howk & Savage (1997, pressure stripping (NGC 134, NGC 4522, NGC 4330, and
1999), who used unsharp mask images in the optical to identify NGC 4388). The galaxies in the Virgo cluster also tend to have
high latitude dust absorption. These include NGC891, less extensive extraplanar H I, but there is no clear difference
NGC4517, NGC4565, NGC5907, NGC4217, and between the UV halos of the Virgo galaxies (NGC 4313,
NGC4157, of which NGC891 and NGC4217 have high NGC 4216, NGC 4330, NGC 4607, NGC 4522, NGC 4217,
latitude dust. The starburst galaxies all have eDIG, soft X-ray NGC 4388, and NGC 5775) and the others. There is also no
halos, extraplanar thermal dust emission (except for apparent connection between the UV halo morphology and H I
NGC 3628), and radio halos. H I emission is also seen around warps, or between the vertical concentration in the H I total
all galaxies in the sample with high resolution maps, although column maps and the diffuse UV images.
in some cases the beamsize makes it unclear how much of the The UV and H I filaments do coincide near the disk in the
emission is truly extraplanar. starburst galaxies NGC253, NGC3628, NGC4666,
Extraplanar diffuse emission: morphology—the extraplanar NGC5775, and NGC3079 (not all shown in Figure 9). The
diffuse emission in normal galaxies shows some of the same UV halo is also brighter and more vertically extended on the
features as the UV halo morphology, although sometimes it is side of the galaxy with the large H I filament in NGC891, and
only detected at lower heights where we cannot measure the a similar phenomenon is seen in NGC3044. However, the
UV halo morphology directly. The eDIG tends to have a large-scale hydrogen filaments seen around galaxies in our
similar morphology as the UV in terms of filaments, sample (such as NGC 55, NGC 3628, NGC 4565, NGC 4631,
concentration, and bilateral symmetry, but it is usually not or M82) are not visible in the UV.
detected at larger heights where the UV halo remains bright. Disk emission—there is good agreement in most normal
The soft X-ray halos correspond reasonably well to the UV galaxies between the projected radial extent (along the disk) of
halo structure in NGC891, NGC2683, and NGC4217, but the bright Hα disk emission and the UV halo above it, which
not in NGC5170. There are fewer normal galaxies with X-ray mostly explains the difference in UV halo concentration among
data than with Hα images, and only four with X-ray halos, so it normal galaxies. The same is true for soft X-rays, radio
is unclear how well the morphology matches the UV. The continuum, and thermal emission from dust in the disk, except
extraplanar 160 μm emission coincides with the UV, but it is for NGC4565 and NGC4522 (Figure 8). In the starburst
12
Figure 8. A comparison of the UV halos to diffuse emission in other bands for normal galaxies. Empty panels indicate where we have no image file, but in some cases
we can compare the morphology from published images (see Table 2). The orange maps show the FUV (or uvw2) contours from Figure 3 while the blue maps show
images at the other wavebands. The thick black contour is the 3σ contour above background for each of these images. The thin black box shows our galaxy region for
PSF-wing contamination. To emphasize the extraplanar emission at 160 μm we also corrected these images for PSF-wing contamination, and the dotted lines show the
galaxy regions used. In some images the map edges are visible as regions of heightened noise, and the contours there should be ignored.
galaxies, the disk emission is less connected to the UV halo and layer does exist, it has a similar morphology to the UV halo and
is instead brightest in the regions where the galactic winds occurs in about the same place. This suggests that the UV halos
occur. arise from some gaseous medium, and this suggestion is
supported by the asymmetric halos in NGC4522 and
4.5. Summary NGC4330 that match the morphology of the stripped gas,
UV halos around luminous spiral galaxies are ubiquitous. In but not the morphology of the stellar disk. The FUV–NUV
normal galaxies, the UV halos tend to form a thick disk. Their colors differ between galaxies, but there are no clear structures
presence is independent of diffuse extraplanar material seen in in the FUV–NUV colors, except perhaps in stripped galaxies.
other wavebands, but in those cases where a diffuse extraplanar The UV halos do appear to be connected to star formation in
13
the disk, as they are brightest over the parts of the disk with 5. ULTRAVIOLET HALO FLUXES
bright Hα, X-ray, or radio emission, but not in a way that
requires ionizing photons or winds to escape the disk. In the prior section we focused on qualitative morphological
UV halos around starburst galaxies are more extended and trends. Here we compare the average halo fluxes and FUV–
have filamentary structures coincident with the galactic winds. NUV colors, as well as the flux as a function of height above
Apart from these filaments, there is little agreement between the the disk, to various metrics for each galaxy. It is especially
extraplanar or disk emission in other wavebands. The winds are important to understand the connection between the structure
visible in the FUV–NUV maps as bluer regions in the UV halo, visible in the UV maps and the average properties because it
and may be superimposed on a disk structure like that in the will not be possible to observe this structure with reasonable
normal galaxies, but the origin of the color difference is not clear. exposure times for more distant galaxies.
14
We measured halo UV fluxes in long boxes parallel to the fluxes, amount of PSF-wing contamination, and uvw2 correc-
midplane of the galaxy, as in HKB14. The length of each box is tion factors are listed in Table 5.
equal to the length of the region containing the galaxy image, In the remainder of this section, we assess the reliability of
and the width of each box is a projected 2 kpc at the distance of the measured fluxes, measure scale heights for the UV halos,
the galaxy. We also reference the height of the box to the and examine the change in color as a function of height. We
projected midplane of the galaxy. For galaxies with lower then search for correlations between the average UV halo
fluxes and galactic properties, and compare our results
inclinations, such as NGC4666, the projected heights are too
to HKB14.
large. The halo fluxes are all measured starting from a projected
height of 2–5 kpc from the midplane in our sample. This 5.1. Reliability of Halo Fluxes
conservative threshold excludes galaxy disk light, but likely
misses halo light at lower latitudes. The galaxy fluxes measured The similarity in the halo morphology between different UV
in the boxes used for PSF-wing correction are reported in filters suggests that our PSF-wing subtraction is accurate, but it
is worth considering the reliability of the measured fluxes when
Table 4.
the PSF-wing contamination can be a large fraction of the total
The halo fluxes are corrected for PSF-wing contamination,
flux. We performed several tests, and conclude that the fluxes
and the uvw2 fluxes are corrected for red leak using a galaxy- are indeed reliable in the sense that the systematic error from
type correction from HKB14 to isolate the true UV flux PSF-wing subtraction is smaller than the statistical error.
(shortward of 3000 Å). We used a correction factor of 0.93, First, the PSF-wing contamination is not very sensitive to the
except for M82 (0.95) and NGC3623 (0.85). The uvw1 filter best-fit exponent because most of the PSF-wing contamination
has a more severe red leak for which this correction is comes from the region where the profile is measured rather than
unreliable, so we report the uncorrected fluxes. The total halo extrapolated. If we extrapolate the PSFs with an exponent of
15
Figure 9. A comparison of the UV halos to diffuse emission in other bands for starburst galaxies. The plot style is the same as in Figure 8.
x=2 in each filter, which is unacceptably shallow (Table 3) contamination, and averaged the uvm2 and uvw1 values.
and inconsistent with the background, the fractional increase in Without the PSF-wing correction, the mean UVOT/GALEX
the PSF-wing contamination is only about 2%, except in the flux ratio is about 1.6 across our sample, whereas after
uvw1 filter where the change is about 8%. correction it is about 1.05. We cannot exactly synthesize the
Second, the NUV filter covers approximately the same NUV filter from the UVOT filters, so a ratio of 1.0 is not
wavelength range as the combined uvm2 and uvw1 filters. expected and this comparison cannot tell us if there are small
Since the PSF-wing contamination is smaller in the NUV, systematic uncertainties. Nonetheless, this exercise suggests
comparing the NUV fluxes to the average of the uvm2 and that the reported fluxes are close to the true flux.
uvw1 can tell us if the PSF-wing contamination is obviously Finally, there are HST observations of the halo of M82 in
too high or low. For galaxies with fluxes in each filter we several UV bands (using the Wide Field Camera 3 UVIS
measured the flux with and without accounting for PSF-wing detector), including the F225W filter that overlaps with the uvw2,
16
NUV, uvm2, and uvw1 filters. The sub-arcsecond angular due to a more complex background that results from the
resolution and small field of view mean that PSF-wing stacking. The starburst stack shows that the scale height
contamination is negligible. The wind is clearly visible in the changes with galactocentric radius, which can be explained by
HST image, and the flux measured in this region is consistent with filaments. The scale heights can also differ between filters, as is
that measured with Swift and GALEX after PSF-wing subtraction the case for M82 and NGC253, but in most galaxies the scale
but not before. For example, in one region (a circle of radius heights measured in different filters agree within the uncer-
15 arcsec centered on α = 09:55:52.22, δ = +69:40:02.39), we tainty. Here we report the value of h measured from the average
measure an AB magnitude of mF225W=16.91±0.04 mag in the radial profile and combining each filter (we allowed A to vary
HST image, muvm2=16.80±0.03 mag in the uvm2 image between filters, but fit for a common h). Table 6 contains the
before PSF-wing correction, and muvm2=16.92±0.03 mag best-fit h and the associated reduced χ2.
after. Similar values are found in the other filters, indicating that Scale heights range from h=1–8 kpc. The mean value of h
the PSF-wing subtraction recovers the true flux to within the for starburst galaxies is h¯ = 3.6 kpc (3.0 kpc when excluding
statistical uncertainty. NGC 4666, which has a lower inclination), and it is h¯ = 3.5
kpc for normal galaxies. There is no dependence on Hα
luminosity, the presence of eDIG, Hubble type, UV halo
5.2. Scale Heights luminosity, or the S/N in the image. Galaxies with stripped gas
UV halos appear to have a morphology like a thick disk, so (NGC 4522, NGC 134, and NGC 4330) have small scale
we measured scale heights (h) based on fitting a function of the heights (h < 2 kpc), but NGC891 and NGC55 have similar
form F (z ) = Ae- z h , where F is the flux, A is a normalization, h values even without stripped gas. Likewise, there is no
and z is the projected height above the midplane. We used the connection between h and the fraction of UV light from
fluxes measured in long boxes because the S/N is only high filaments. The galactic or halo properties that seem to be
enough in a few galaxies to measure scale heights at different connected to other morphological indicators of UV halos do
radii. To see how the aggregate behaves, we fit scale heights at not predict the scale height.
several galactocentric radii in the stacked quadrant images. In several galaxies, h varies between filters. For example, in
Figure 5 shows the scale height as a function of galactocentric NGC253 the joint fit to all filters results in h=2.2 but with a
radius, along with the 1σ error bars, for the normal and reduced χ2=38. The FUV profile is marginally fit for
starburst galaxies. In the starburst galaxies the source mask h=1.8 kpc (reduced cn2 = 2), but the NUV profile cannot
near R=0 for NGC4631 is taken into account. In the normal be fit with a single exponential profile (h = 2.6 kpc with
galaxy stack the scale height within R25 is consistent with being reduced χ2 = 43). A double exponential profile is a good fit
constant, with a possible increase with larger radii that may be with an inner component (h ≈ 1 kpc) and an outer component
17
Table 4
Galaxy UV Fluxes
Name FUV NUV uvw2 uvm2 uvw1 uvw2 Factor

(1) (2) (3) (4) (5) (6) (7)
Starburst Galaxies
NGC 0253 105.20±0.01 200.78±0.02 L L L 0.93

M82 3.67±0.04 17.30±0.08 14.47±0.09 13.49±0.07 37.96±0.04 0.95
NGC 4631 92.80±0.06 132.3±0.1 108.4±0.2 121.3±0.3 132.6±0.4 0.93
NGC 3628 5.45±0.02 14.6±0.1 11.6±0.1 11.9±0.2 28.3±0.9 0.93
NGC 4666 5.69±0.04 10.20±0.04 8.31±0.07 8.20±0.09 13.95±0.08 0.93
NGC 3079 10.39±0.01 16.29±0.02 13.96±0.02 16.12±0.04 20.9±0.1 0.93
NGC 5775 2.47±0.01 4.48±0.03 4.38±0.01 4.13±0.02 7.36±0.03 0.93
NGC 4388 3.16±0.01 6.50±0.01 5.68±0.01 6.36±0.01 9.87±0.02 0.93
Normal Galaxies
NGC 0055 294.56±0.01 441.60±0.03 L L L 0.93

NGC 0891 2.85±0.06 5.7±0.2 5.5±0.2 4.6±0.3 10.77±0.05 0.93
NGC 2683 6.36±0.02 12.35±0.04 10.27±0.04 10.23±0.09 19.9±0.3 0.93
NGC 4517 6.39±0.02 10.27±0.03 L L L 0.93
NGC 4565 L 15.45±0.01 L L L 0.93
NGC 4096 7.26±0.03 12.80±0.04 10.39±0.03 11.17±0.03 L 0.93
NGC 4313 0.24±0.01 1.00±0.01 L L L 0.93
NGC 3623 3.25±0.04 9.58±0.05 7.60±0.03 8.04±0.02 20.66±0.05 0.85
NGC 5907 7.99±0.03 12.96±0.05 10.99±0.03 11.58±0.06 16.75±0.04 0.93
NGC 4216 1.42±0.02 3.70±0.02 L L L 0.93
NGC 4607 0.22±0.01 0.60±0.01 L L L 0.93
NGC 4522 2.57±0.01 4.03±0.02 L L L 0.93
NGC 0134 L L 10.28±0.05 10.46±0.07 17.4±0.2 0.93
NGC 4157 3.13±0.02 9.60±0.04 L L L 0.93
ESO358-063 1.62±0.01 3.30±0.02 L L L 0.93
NGC 4217 0.74±0.03 1.50±0.03 L L L 0.93
NGC 4330 1.41±0.01 2.27±0.01 1.87±0.02 1.99±0.02 2.64±0.03 0.93
NGC 3044 5.77±0.02 9.30±0.02 6.98±0.04 7.74±0.03 9.56±0.03 0.93
NGC 5170 2.40±0.07 5.06±0.04 L L L 0.93
Note. The galaxy fluxes are measured in boxes as described in Section 5. The uvw2 scale factor is based on the galaxy type as described in HKB14.
(h ≈ 3 kpc), so it is possible that the NUV light has a compact markedly differs across the midplane in a way that other filters
component and a component similar to the FUV halo. Figure 4 do not. The most likely explanation is that the wind is brightest
shows that the NUV light follows the disk much better than the in the FUV, as seen in Figure 7. Most galaxies have
FUV light, suggesting a possible difference in physical origin. approximately symmetric flux profiles, despite differences in
A similar difference is found in M82, where the joint fit the visible morphology across the midplane (e.g., Figure 4).
produces h=2.7 kpc (cn2 = 32) and the best-fit values of h in The galaxies where there is a notable difference in multiple
each filter are: h=4.9, 3.0, 2.9, 3.2, and 2.5 kpc for the FUV, filters include NGC4666, NGC4522, NGC4330, and
uvw2, uvm2, NUV, and uvw1 filters, respectively (with NGC134. The latter three galaxies have an obvious asymmetry
corresponding cn2 = 1.7, 4.4, 0.8, 2.9, and 12). As in due to stripped gas (Figure 3). NGC4666 is the least inclined
NGC253, the FUV halo is more extended, and the redder galaxy in the sample, and the asymmetry in scale heights may
pertain to inclination and projection effects, as mentioned
filters cannot be adequately described by single exponential
in HKB14.
models. On the other hand, in NGC55 (joint h = 1.1 kpc and
cn2 = 4.2 ) the FUV halo has a smaller scale height than the
NUV (h ≈ 0.5 kpc compared to 1.3 kpc). It is tempting to 5.3. FUV–NUV Color with Height
ascribe the difference in behavior to the presence of galactic We measured the correlation between the FUV–NUV color
winds, and indeed NGC3079, NGC4631, and NGC4666 do in each flux measurement bin and the projected height of the
have shallower FUV than NUV profiles, but the S/N is bin for each galaxy with data in both filters. Most galaxies have
generally too low in the normal galaxies to determine if this is a no significant correlation (for a threshold of p = 0.05), and the
clear difference between them. exceptions are all starbursts. In NGC253, M82, NGC4631,
We also searched for different scale heights across the NGC3628, NGC3079, and NGC4388, the FUV–NUV color
midplane. In many cases, one filter has a different profile across decreases with height. This is consistent with the FUV–NUV
the midplane while the others do not, but these can usually be color maps (Figure 7) and suggests that the UV halos comprise
explained by a small background gradient across the region of two components.
the galaxy (e.g., due to Galactic cirrus, which differs in strength To improve the signal, we measured the FUV–NUV color as
between filters). The exception is M82, where the FUV a function of height in composites of the normal, starburst, and
18
Table 5
Halo UV Fluxes
Halo Flux (mJy) PSF-Wing Contamination (mJy) uvw2 Factor

Name FUV NUV uvw2 uvm2 uvw1 FUV NUV uvw2 uvm2 uvw1
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Starburst Galaxies
NGC 0253 5.43±0.09 17.2±0.2 L L L 4.89 0.60 L L L 0.93

M82 7.0±0.1 15.7±0.2 14.4±0.2 13.3±0.2 25.9±0.1 0.29 0.53 0.69 0.54 2.94 0.95
NGC 4631 2.74±0.07 5.3±0.2 5.0±0.3 5.0±0.4 6.20±0.5 5.75 0.99 3.16 3.01 6.39 0.93
NGC 3628 0.46±0.01 1.92±0.09 1.4±0.1 1.2±0.1 3.3±0.4 0.20 0.11 0.21 0.19 0.80 0.93
NGC 4666 0.23±0.02 0.51±0.02 0.49±0.03 0.36±0.04 0.48±0.04 0.16 0.06 0.13 0.11 0.39 0.93
NGC 3079 0.42±0.01 0.65±0.02 0.65±0.05 0.63±0.04 0.5±0.1 0.53 0.23 0.43 0.41 1.15 0.93
NGC 5775 0.23±0.01 0.39±0.02 0.49±0.01 0.40±0.01 0.71±0.03 0.13 0.11 0.17 0.14 0.45 0.93
NGC 4388 0.12±0.01 0.37±0.01 0.25±0.01 0.29±0.01 0.90±0.02 0.14 0.08 0.16 0.14 0.49 0.93
Normal Galaxies
NGC 0891 0.58±0.05 0.8±0.2 0.8±0.1 1.0±0.3 1.3±0.2 0.15 0.05 0.15 0.10 0.47 0.93
NGC 2683 0.30±0.02 0.88±0.04 0.84±0.04 0.5±0.1 1.6±0.4 0.29 0.11 0.28 0.22 0.94 0.93
NGC 4517 0.23±0.02 0.9±0.4 L L L 0.33 0.12 L L L 0.93
NGC 4565 L 1.08±0.09 L L L L 0.17 L L L 0.93
NGC 4096 0.10±0.02 0.48±0.04 0.54±0.03 0.41±0.02 L 0.34 0.18 0.31 0.27 L 0.93
NGC 4313 0.05±0.01 0.12±0.02 L L L 0.01 0.03 L L L 0.93
NGC 3623 0.02±0.02 0.21±0.03 0.19±0.02 0.17±0.01 0.47±0.03 0.06 0.01 0.07 0.06 0.37 0.85
NGC 5907 0.22±0.02 0.43±0.04 0.39±0.02 0.43±0.04 0.42±0.03 0.39 0.10 0.29 0.25 0.77 0.93
NGC 4216 0.08±0.01 0.34±0.02 L L L 0.07 0.08 L L L 0.93
NGC 4607 0.02±0.01 0.07±0.01 L L L 0.01 0.03 L L L 0.93
NGC 4522 0.15±0.01 0.25±0.02 L L L 0.11 0.09 L L L 0.93
NGC 0134 L L 0.44±0.03 0.24±0.05 0.6±0.1 L 0.26 0.21 0.75 0.93
NGC 4157 0.15±0.02 0.31±0.01 L L L 0.13 0.08 L L L 0.93
ESO358-063 0.06±0.01 0.19±0.02 L L L 0.06 0.03 L L L 0.93
NGC 4217 0.17±0.03 0.50±0.03 L L L 0.04 0.06 L L L 0.93
NGC 4330 0.11±0.01 0.17±0.01 0.20±0.03 0.15±0.02 0.26±0.05 0.08 0.16 0.12 0.11 0.26 0.93
NGC 3044 0.11±0.02 0.13±0.01 0.09±0.04 0.10±0.03 0.02±0.05 0.31 0.32 0.29 0.28 0.69 0.93
NGC 5170 0.04±0.05 0.67±0.05 L L L 0.08 0.09 L L L 0.93
Note. The halo fluxes are measured as described in Section 5. The uvw2 scale factor is based on the galaxy type as described in HKB14. The total flux measured on the
CCD is the sum of the inferred astrophysical flux (“halo flux”) and the PSF-wing contamination.
stripped galaxies (of which NGC 4522 and NGC 4330 have luminosity of the galaxy (LHα), the stellar mass (M*), the
GALEX data). The composite measurements are shown in rotation velocity (vrot), the morphological type code (T; de
Figure 10. The data were binned in boxes 2 kpc wide, with the Vaucouleurs et al. 1991), the inclination (i), and the distance
first box having a central height of 3 kpc from the midplane. from Earth (d).
The projected distances from the midplane for each measure- LHα, vrot, T, i, and d come from NED and HyperLeda values.
ment were used to assign each measurement to the appropriate Some of the Hα fluxes include some flux from the neighboring
bin, and the galaxies were projected to a common distance of [N II] line. We compute M* from the K-band luminosity from
20 Mpc for the purpose of computing the magnitudes. Figure 10 the 2MASS Extended Source Catalog and the mass-to-light
reinforces the conclusions from the individual galaxies: the ratio from Bell & de Jong (2001), for which we use B−V
starburst halos become bluer with height, the normal galaxies colors from HyperLeda. The SFR is based on the relation from
have an approximately constant color, and the stripped galaxies Kennicutt (1998), which uses far-infrared (FIR) fluxes
are strikingly blue near the disk and similar to the normal measured with the IRAS satellite. Fluxes exist for all but one
galaxies at larger heights. However, some variation in the color galaxy (NGC 4217). The values we use are reported in Table 1.
with height is seen in a few individual normal galaxies. More We estimated Lgal in each filter by “de-reddening” the
data are necessary to determine if this is similar to what is seen measured flux following Buat et al. (1999), who found a
in the starbursts. relation for the UV extinction based on the FIR and measured
UV fluxes. The relation is best calibrated for the FUV and
uvm2 filters, so we focus on these filters. We further correct Lgal
5.4. Correlation Analysis for Average Properties for the light that would escape the galaxy assuming an
We compared the total halo fluxes, FUV–NUV colors, and extinction out of the disk of 0.5 mag in the B band
the scale heights to the following galaxy parameters: the UV (Calzetti 2001). We convert this to the extinction in the FUV
luminosity of the galaxy (Lgal, corrected for internal extinction), and uvm2 filters using a modified Calzetti et al. (2000) law
the star formation rate (SFR) and specific SFR (sSFR), the Hα from Kriek & Conroy (2013), who derived extinction curves
19
The Astrophysical Journal, 833:58 (29pp), 2016 December 10
Table 6
Diffuse Halos
Extraplanar Diffuse? Lhalo Lgal Lhalo/Lgal

Name FUV–NUV Scale Height cn2 FFIR A0.16 A0.2
(mag) (kpc) (Hα) (X-ray) (160μm) (GHz) (mag) (mag) (FUV) (uvm2) (FUV) (uvm2) (FUV) (uvm2)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17)
Starbursts
NGC 0253 1.26±0.02 2.2±0.1 38 Y Y Y Y 67.6 3.6 L 6.9 L 1448.2 L 0.005 L

M82 0.87±0.02 2.7±0.1 32 Y Y Y Y 86.8 7.5 6.2 13.0 24.6 2621.6 3151.3 0.005 0.008
NGC 4631 0.73±0.03 1.8±0.1 1.3 Y Y Y Y 6.4 1.5 1.2 11.9 21.8 612.7 648.2 0.019 0.035
NGC 3628 1.55±0.08 3.9±0.2 1.8 Y Y N N 5.7 4.1 3.0 6.5 17.2 1347.4 1128.5 0.005 0.016
NGC 4666 0.90±0.08 7.6±0.4 0.3 Y Y Y Y 3.9 3.7 3.0 8.2 12.9 2333.3 1932.9 0.003 0.007
NGC 3079 0.35±0.01 3.4±0.1 1.6 Y Y Y Y 4.9 3.3 2.6 18.7 28.0 3772.2 3214.7 0.005 0.009
NGC 5775 0.59±0.04 4.5±0.3 2.6 Y Y L Y 2.3 4.0 3.2 11.4 19.8 1887.0 1563.7 0.006 0.013
NGC 4388 1.02±0.06 2.7±0.1 3.5 Y Y Y Y 1.1 3.0 2.1 6.0 14.6 945.9 878.7 0.006 0.017
Normal Spirals
NGC 0055 2.8±0.6 1.1±0.1 4.2 Y N Y L 4.0 0.5 L 0.5 L 65.7 L 0.007 L
NGC 0891 0.33±0.09 1.5±0.2 0.8 Y Y Y Y 6.8 5.0 4.3 6.9 11.4 1088.3 1043.1 0.006 0.010
NGC 2683 2.0±0.4 2.7±0.2 0.9 N N L N 1.1 2.3 1.7 3.6 5.8 203.6 219.7 0.018 0.025
NGC 4517 1.5±0.2 3.4±0.2 1.9 L L Y L 0.7 1.9 L 3.1 L 152.2 L 0.020 L
NGC 4565 L 2.0±0.3 0.7 N N N N 1.2 L L L L L L L L
NGC 4096 1.8±0.3 4.3±0.3 0.7 N L Y N 0.9 2.0 1.5 1.9 7.9 277.6 312.3 0.007 0.024
20
NGC 4313 1.0±0.3 6.6±3.3 2.7 L L L L 0.2 3.9 L 1.3 L 73.7 L 0.021
NGC 3623 3.±3. 4.±1. 3 N N L N 0.6 2.3 1.4 0.4 3.3 169.9 209.4 0.002 0.015
NGC 5907 0.74±0.09 6.4±0.2 1.3 N N Y N 1.7 2.5 2.0 7.1 13.8 816.6 829.1 0.009 0.016
NGC 4216 1.6±0.3 3.4±0.3 0.9 N L N N 0.3 2.6 L 2.7 L 160.1 L 0.017
NGC 4607 1.2±0.4 3.7±1.3 1.0 L L N L 0.4 4.7 L 0.8 L 202.1 L 0.004 L
NGC 4522 0.55±0.04 1.9±0.1 1.8 Y L Y L 0.2 1.6 L 6.0 L 139.4 L 0.043 L
NGC 0134 L 1.9±0.3 0.7 L L L L 2.8 2.5 L 10.1 L 1548.1 L L
NGC 4157 0.8±0.1 4.7±0.5 2.1 N L L Y 2.0 3.6 L 6.3 L 1171.6 L 0.005 L
ESO358-063 1.2±0.1 5.3±0.9 0.1 L L L L 0.5 2.8 L 2.6 L 296.9 L 0.009 L
NGC 4217 1.2±0.2 3.1±0.4 0.1 Y Y Y Y L L L 7.6 L L L L L
NGC 4330 0.51±0.03 1.1±0.1 1.6 L L L L 0.1 1.7 1.4 5.1 6.9 101.7 116.5 0.050 0.055
NGC 3044 0.2±0.1 3.8±0.7 0.8 Y Y L Y 1.1 2.4 1.9 6.7 6.0 998.1 1011.2 0.007 0.006
NGC 5170 3.±2. 7.9±0.5 0.8 N N L L 0.2 1.5 L 3.6 L 269.1 L 0.013 L
Hodges-Kluck, Cafmeyer, & Bregman

Note. Columns (1) Name, (2) FUV–NUV color averaged over the halo, (3–4) Best-fit scale height to combined data from all wavebands and reduced χ2, (5–8) Is there extraplanar diffuse emission?, (9) FIR flux used to
determine galactic extinction in units of 10−9 erg s−1 cm−2, (10–11) Estimated extinction in the FUV and uvm2 bands from Buat et al. (1999), (12–13) Halo luminosity density in units of 1025erg s−1 Hz−1, (13–14) De-
reddened galaxy luminosity density in units of 1025erg s−1 Hz−1, (15–16) Halo-to-galaxy luminosity ratio using de-reddened galaxy luminosity.
References. The presence of extraplanar diffuse emission was determined from the literature. Hα: Rossa & Dettmar (2003a), X-ray: Strickland et al. (2004), Tüllmann et al. (2006), Li & Wang (2013), Radio: Wiegert
et al. (2015).
Table 7
UV Halo Correlations
FUV–
Quantity Lhalo Lhalo/Lgal h NUV
(FUV) (uvm2) (FUV) (uvm2) (kpc) (mag)
(1) (2) (3) (4) (5) (6) (7)
Lgal,FUV 0.87a - N - N 0.41a
Lgal,uvm2 - 0.79a - N N N
h N N N N - N
Lgal,Hα 0.62a N N N N N
SFR(IR) 0.82a 0.77a 0.42 0.59 N N
M* N N N N N N
SFR 0.64a 0.59 N N N 0.48
(IR)/M*
vrot N N N N N N
T N 0.58 N N N N
Figure 10. FUV–NUV color as a function of height for composites of the i N N N N N N
normal, starburst, and stripped galaxies in our sample. d N N N N 0.47 N
Note. Correlations between halo quantities and galaxy and halo properties. For
significant correlations (p < 0.05) the Spearman ranked correlation coefficient
for composite galaxy SEDs. Following their scheme, we adopt is given. Non-significant correlations are marked by an “N.” A dash (“-”)
an attenuation slope of δ=−0.2 for the starbursts and δ=0.0 indicates no measurement. Column (1) Quantity, (2–3) Halo luminosity, (4–5)
Ratio of halo luminosity to de-reddened galaxy luminosity, (6) Scale height of
for the normal galaxies, and a 2175 Å bump strength of
the UV halo, (7) Average FUV–NUV color in the halo. See text for more a
Eb=0.5 or 1.0 for starbursts and normal galaxies, respec- more detailed description.
tively. These are both more similar to the Calzetti et al. (2000) a
These correlations survive the Benjamini–Hochberg test with α=0.05 for
law than the traditional Milky Way extinction law toward 61 tests.
(Weingartner & Draine 2001). This leads to AFUV/AB=2.05
and Auvm2/AB=1.96 for the starburst galaxies, whereas AFUV/
AB=2.5 and Auvm2/AB=2.2 for the normal galaxies. These
parameters are appropriate for star-forming galaxies, but the Lhalo is also correlated with the FUV–NUV color for the
analysis below is insensitive to the exact δ or Eb, or whether FUV band (R = 0.52, p = 0.00076), but not uvm2 (p = 0.11).
Lgal includes attenuation along the minor axis. The measured The FUV–NUV color is significantly correlated with the
Lgal, the estimated extinction, and the corrected values are specific SFR (R = 0.48, p = 0.028) but not the SFR
given in Table 6. (p = 0.076), nor with Lgal in either FUV or uvm2 (p = 0.051
We used Spearman’s rank correlation with a significance and p = 0.38, respectively). There are no other significant
threshold of p=0.05 to search for correlations between galaxy correlations between FUV–NUV and other galaxy properties.
In addition to the FUV Lhalo, the FUV–NUV color is
and halo properties. A summary of the findings is given in
correlated with the FUV Lgal (R = 0.41, p = 0.048). It is not
Table 7. Here we describe the results and give the correlation
correlated with other halo luminosities. FUV–NUV is also
coefficient (R) and p-value for correlations with p<0.05. As
marginally correlated with sSFR (R = 0.48, p = 0.03), but the
we tested for correlations between a range of measurements, we
correlation with SFR is not significant. There are no other
adopted the Benjamini & Hochberg (1995) procedure to
significant correlations between FUV–NUV and the galaxy
control for false positives, also using a threshold of α=0.05.
properties.
There were 61 total comparisons. Correlations with p-values We also compared Lhalo/Lgal to the galaxy properties for the
that remain significant for the revised threshold are noted in FUV and uvm2 filters. Lhalo/Lgal is correlated with the SFR
Table 7 and discussed below. (R = 0.42 and p = 0.042 for the FUV, and R = 0.59 with
The halo luminosity Lhalo is strongly correlated with Lgal, p = 0.027 for uvm2). There are no other significant correlations
with R=0.87 and p<10−8 in FUV and R=0.79 and with other properties.
p=8.3×10−4 in uvm2. It is also correlated with the SFR in The scale height h is not correlated with any galaxy or halo
both (R = 0.82 and p = 8.5 × 10−7 for the FUV and R = 0.77 property except d, where R=0.47 and p=0.012. This is
with p = 0.0012 in uvm2), and with the specific SFR (R = 0.64 almost entirely due to NGC5170 (h = 7.9 kpc), which is near
and p = 0.0016 in FUV and R = 0.59 and p = 0.033 in uvm2). the distance cutoff and has an unusually large value.
LHα is closely related to the SFR, but whereas there is a Applying the Benjamani–Hochberg procedure for α=0.05
significant correlation between Lhalo and LHα in the FUV and 61 tests, we find that the following correlations are
(R = 0.62, p = 0.0041) there is no significant correlation for significant: in both FUV and uvm2 Lhalo is correlated with Lgal
uvm2 (p = 0.12). Lhalo is correlated with morphological type for and with the SFR, and Lhalo is correlated with FUV–NUV,
uvm2 (R = 0.58, p = 0.030) but not for the FUV (p = 0.12). specific SFR, and LHα in the FUV only. The other correlations
Overall, the p-values are smaller for the FUV, possibly because with p<0.05 described above are not significant at this
there are 24 FUV measurements and 15 uvm2 measurements. threshold.
There is no other significant correlation between Lhalo and the In summary, Lhalo is related to Lgal and some the metrics of
other parameters in either filter. star formation, as well as the halo color. However, the scale
21
height has no (strong) dependence on the halo or intrinsic possible that any astrophysical UV halo exists primarily below
galaxy properties. 2 kpc. Based on these results and the detections in our present
sample, UV halos are ubiquitous but may not be universal (or
universally detectable with current instruments).
5.5. Average Properties and Morphology
For most galaxies in the GALEX and Swift archive it is not 6. DISCUSSION
possible to examine the morphology in detail because of low
S/N or large distance. Thus, we compare the average properties 6.1. The Diffuse UV Light is Probably a Reflection Nebula
Lhalo, FUV–NUV, and h to the UV maps in Figures 3 and 4. Prior studies have argued that the UV halos are eRN on the
Overall, we find that averaging obscures the presence of halo basis that the light is too bright to come from shock-heated or
components such as winds, filaments, or bright emission near photoionized gas (Hoopes et al. 2005), that line emission from
the disk, so that these properties are not useful metrics of the starburst halos is polarized in a way that is consistent with dust
halo morphology. scattering (Yoshida et al. 2011), that it can be successfully
Apart from the clear divide between normal and starburst modeled by Monte Carlo radiative transfer (MCRT) scattering
galaxies, Lhalo appears unrelated to the halo morphology, models (Seon et al. 2014; Shinn & Seon 2015), that it is too
except in stripped galaxies. For example, NGC5907 has a very blue to originate in the stellar halo HKB14, and that they
patchy, low S/N halo in Figure 3, but its Lhalo is among the coincide with eDIG, winds, and other outflow tracers.
highest of the normal galaxies. Since we know that Lhalo is Our results further support this scenario:
uncorrelated with h it does not provide information about the
halo structure. Likewise, the most spectacular UV halos around 1. The UV halos are truly diffuse, as determined by
starburst galaxies (M82, NGC 253, NGC 4631, and NGC 3079) comparison with existing HST data, and they also
have similar Lhalo. The stripped galaxies (NGC 134, trace diffuse filamentary morphology seen at other
NGC 4522, and NGC 4330) have unusually high Lhalo/Lgal wavelengths.
(5%–10% instead of the typical 0.5%–2%). As discussed 2. UV halos are a broadband phenomenon, with similar
earlier, this is largely attributable to extraplanar star formation. morphology seen from 1500 to 2600 Å.
We also find no connection between the halo morphology and 3. Lhalo is strongly correlated with Lgal after de-reddening
the average FUV–NUV color, except insofar as bluer halos are the galaxy flux, and it is strongly correlated with the SFR.
more structured in our sample because they predominantly belong 4. Extraplanar UV light is most visible above regions of
to starburst galaxies. Likewise, the scale height appears to be active star formation as seen in Hα, X-ray, and radio
independent of radial concentration, although in starburst or continuum data, but the presence or prominence of the
stripped galaxies the scale height can change with galactocentric UV halo does not depend on the presence of extraplanar
radius. The stripped galaxies cannot be distinguished from the rest Hα, X-rays, or radio continuum. This rules out an
of the sample by their average FUV–NUV color and are only emission nebula as the source.
barely distinguishable by scale height; it is not likely that these 5. Starburst galaxies have more luminous UV halos, but not
metrics could identify stripped galaxies in a wider sample. higher Lhalo/Lgal values or scale heights.
6. Starburst winds are visible through their smaller FUV–
NUV colors, which appear to be superimposed on a thick
5.6. Differences with HKB14 disk similar to (although brighter than) that around
In HKB14 we measured fluxes without the PSF-wing normal galaxies.
correction and used these fluxes to find scale heights for several 7. The UV halo flux, color, and scale height are not
galaxies and measure correlations between some of the same significantly correlated with galaxy properties that are not
quantities as above. The PSF-wing contamination is strongest near closely connected to star formation.
the disk, so we might expect h to increase when the spurious flux The strong connection between Lhalo and Lgal and the weak
is removed, but this is not generally the case. However, the connection between Lhalo/Lgal and the presence of outflows
correlations that we find between Lhalo and Lgal, the Hα strongly suggests that UV halos are eRN. This also explains
luminosity, and the FIR SFR are stronger after Lhalo is corrected their broadband visibility and ubiquity, as non-ionizing UV
for the PSF-wing contamination. For example, in the FUV the photons can escape even if the disk is not porous. If UV halos
correlation between Lhalo+LPSF contam and Lgal is R=0.41 are eRN, then they indicate that dust is widespread in the halos
whereas between Lhalo and Lgal it is R=0.87. The lower fluxes of spiral galaxies of all types. This implies that dust is long-
also change the SEDs, but the only trend reported in HKB14 that lived in galaxy halos to the extent that it exists around galaxies
is truly artificial is the uniform rise in the FUV–NUV color with without strong outflows and has a similar scale height as the
height; we find in most galaxies no change with height. dust around starburst galaxies. In the remainder of this section,
Shinn & Seon (2015) investigated the role of PSF-wing we assume that the UV halos are eRN and examine some basic
contamination in several galaxies included in HKB14, and properties (however, we defer a detailed analysis of the UV
found that in two cases (NGC 24 and IC 5249) the UV halo that halo SEDs to Paper III). We then discuss the results in the
we reported appears to be entirely artificial. They used the context of other work on extraplanar dust.
GALEX images exclusively, and extrapolated the PSF wings
from the PSF provided by the GALEX calibration team. Using
6.2. Dust Mass
the PSFs described in Section 3, we agree with their findings
for the GALEX images. For the considerably deeper Swift We estimate the dust mass using MCRT models based on the
images, we find a faint UV halo around NGC24 (which is not model described in Wood et al. (2001) and Whitney (2011),
edge-on) and no UV halo above 2 kpc around IC5249. which incorporates the Henyey–Greenstein functions for the
IC5249 is seen edge-on and has a very thin optical disk, so it is angular dependence of dust scattering. These models are meant
22
Figure 11. The best-fit MCRT model to NGC4631 for Milky Way-like dust. Top left: MCRT image at 2000 Å with 3 and 6σ contours overlaid, along with flux
measurement boxes. The average noise level from the uvw2 image was added to create the contours and show the “observable” extent of the halo. Top right: measured
uvw2 fluxes from NGC4631 compared to values measured from the MCRT model in the boxes at left. Fluxes in both cases are normalized to the measured galaxy
flux. Bottom left: model H I contours overlaid on the MCRT image, which is rotated to the position angle of NGC4631. Bottom right: measured H I column densities
overlaid on our H I model. The thick black line shows the profile along the minor axis, while the thin line shows the major axis.
to be first-order estimates and are not as carefully constructed exponential disk that represents the halo. Each exponential disk
as other MCRT models of eRN such as in Shinn & Seon (2015) takes the form r (R, z ) = r0 e-R R0 e- z z0 , where R is the
or Baes & Viaene (2016). galactocentric radius in the plane and z is vertical height above
In principle, the dust mass can be measured directly from the the plane. We constrain ρ0, R0 and z0 by using high resolution
Lhalo/Lgal ratio, since H I maps, so we only model galaxies where we have such maps
L halo, n = L gal, n (1 - e-tn vn ) (1 ) in hand: NGC891, NGC4631, NGC5775, and NGC5907.
R0 in the halo is unconstrained, so we tie it to the value of R0 in
where ϖν is the scattering albedo and tn = 1.086Ndust snext is the the disk. The spiral arms are described by a polar equation
optical depth. snext is the extinction cross-section and Ndust the r=ae bθ, where a is the normalization and b the rate of growth
column. The extinction through the halo is low (τν = 1), so a (the parametric equations for a Cartesian grid are x = r cos q
single-scattering approximation is reasonable. As described and y = r sin q ). We choose a=7 and b = 0.28 and compute
in HKB14, snext and ϖν can be determined from the shape of x and y for q Î [0, 3p ] with a minimum radius R=3 kpc. To
the SED, so the normalization is related to a characteristic add thickness, these spiral arms are convolved with a 3D
column density from the disk through the halo. However, the Gaussian kernel with σ=8 pixels and an amplitude of 0.1 ρ0,
measured flux at a given height above the disk cannot be which is added to the underlying exponential disk. We fixed the
straightforwardly interpreted as a dust column because the light parameters based on the extent of typical spiral arms, but for an
source is not behind the dust. Thus, one must also adopt a edge-on viewing angle our results are insensitive to a wide
geometric model for the halo dust and the disk emission, for range of a, b, or spiral arm width. The lower panels of
which we use MCRT. Figure 11 show the model for NGC4631; we do not attempt to
The MCRT model needs input distributions for the emission reproduce large-scale filaments or warps.
and the scattering/absorbing medium, as well as the viewing The light source consists of a very thin exponential disk with
angle and the scattering cross-section (snext ´ vn ). the same scale length as the thin gas disk and light from the
We assume that the dust is embedded in the neutral medium midplane in the spiral arms, since we assume that UV light
and model the gas density using several components: a thin comes from young stars near the midplane and in the arms. Our
exponential disk, simple logarithmic spiral arms, and a thick results are not sensitive to the shape or the location of the spiral
23
Table 8
MCRT Models and 160 μm Masses
Thick Disk Milky Way Dust SMC Dust

Name R0 z0 G/D cn2 Mext G/D cn2 Mext M160 μm
(kpc) (kpc) (106 Me) (106 Me) (106 Me)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
+80 +120 +2
NGC 891 12 1.9 60-20 1.0 6±3 50-10 0.9 7-5 3.2
+150 +0.3 +400
NGC 4631 8 2.3 450- 50 4.6 1.2-0.1 700- 200 4.1 0.7±0.3 0.7
+50 +0.6 +100 +0.4
NGC 5775 12 3.0 250- 50 9.9 4.0- 0.8 400- 50 9.3 2.3- 0.5 L
+250 +1.5 +200 +3.1
NGC 5907 17 3.0 400- 150 1.1 2.5- 0.8 300- 150 1.4 3.3- 1.3 6.1
Note. Columns (1) Name, (2–3) Scale lengths for the thick exponential disk from matching H I profiles, (4–6) Best-fit D/G, reduced χ2, and implied extraplanar dust
mass above 2 kpc for models with Milky Way dust. The error bars are statistical from the MCRT fitting only and underestimate the true uncertainty. (7–9) Best-fit
D/G, reduced χ2, and extraplanar dust mass for SMC dust. The fits were performed using a dust opacity and scattering albedo at 2000 Å and uvw2 data. See text
for details. (10) Lower bound to dust mass above 2 kpc from 160 μm fluxes. See text for details.
arms because we measure the halo fluxes by summing over R The morphology of the observed halos is more complex than
(described below). the symmetric, disky morphology that we obtain by construc-
PaperIII will be devoted to modeling the SED and tion, but since Lhalo and the scale height h are not
determining the dust size and composition, which will yield morphological indicators (Section 5), the MCRT models are
snext and ϖν. Here we adopt a simpler approach by using the reasonable proxies for obtaining Mdust (Seon et al. 2014). The
values from two dust models from Weingartner & Draine top right panel shows the best-fit model compared to the
+150
(2001): dust in the bar of the Small Magellanic Cloud (SMC), measurements. The best-fit G/D values are G D = 450- 50 for
+400
and dust in the Milky Way for sightlines where the extinction Milky Way dust and G D = 700-100 for SMC dust, which
law RV=3.1. The choice of these models is motivated by the implies a dust mass above 2 kpc of Mdust = 1.2- +0.3
0.1 ´ 10 M
6
observation that the halo dust at larger radii appears to have an for Milky Way dust and Mdust=0.7±0.3×106 Me for
SMC-like extinction curve (Ménard et al. 2010), and that many SMC dust. We emphasize that we do not know whether SMC
galaxy disks have extinction curves similar to the Milky Way. or Milky Way dust is closer to the true halo composition, but
As we shall see, the inferred dust masses for each model are the similarity in masses suggests that Mdust is insensitive to the
similar. dust model for a plausible range of models. The bottom panels
The density and emission model is gridded on a of Figure 11 show the H I model compared to the data.
201×201×201 cube with a pixel scale of 2.5 pixels kpc−1, In Table 8 we give the best-fit G/D, χ2, halo component
and we model the galaxy at 2000 Å only (where the corrected parameters, and dust mass above ∣z∣ > 2 kpc for the four
Lgal is most reliable). The viewing angle is matched to the galaxies we considered. We expect the best-fit G/D to be
inclination of each galaxy, and we apply the dust model by similar between filters, but with the current data this is not a
assuming that the extinction within the disk follows the Kriek useful test because the true extinction curve is not known, and
& Conroy (2013) prescription with δ=0.0 and Eb=1.0, with changes in G/D are degenerate with changes in the incident
a gas-to-dust ratio (G/D) of 100, which is typical for spiral spectrum. We caution that there could be systematic shifts in
disks. Then, for a given halo G/D we simulate the light G/D due to our choice of geometric model, the assumption that
scattered into the line of sight above 2 kpc from the midplane the dust is hosted by the neutral medium, the G/D adopted in
and compare to the observations. We repeated this procedure the disk, and the scattering recipe in the MCRT code. Even if
for a grid of G D Î [50, 1500] with spacing of 10 between all of these are correct or unimportant, we expect the MCRT
G/D=50–100, and a spacing of 50 thereafter. The 1σ Mdust values to be lower limits to the halo dust in this region
acceptable range shown in Table 8 is based on Δχ2 on this because shallower profiles are allowed by the H I data and
grid. The best-fit halo G/D then implies a dust mass. For each could not be constrained by our UV measurements.
simulation, we use 107 scattered photons (we found for several Bearing these caveats in mind, the MCRT Mdust masses are
test cases that results do not differ when using 108 photons). consistent with extrapolating the spherically symmetric Ménard
For each G/D we simulated an image and measured the halo et al. (2010) radial profile (measured between 20 and 1000 kpc)
light using long boxes parallel to the midplane that are to 2–10 kpc, when we scale their average profile based on the
analogous to those used to measure real UV halos. This binning ratio of the mass of each of the four galaxies considered here to
smooths over differences between spiral arm models that are their average galaxy. This suggests that there is a smooth
not well constrained by the data. We normalized the real and density transition from the disk to the halo, at least in
simulated fluxes to the projected galaxy luminosity, and found projection.
the best-fit G/D using the χ2 statistic. The best fit is often not a
good fit, but the simulated halos reasonably reproduce the
6.3. 160 μm Dust Mass
observed scale heights and Lhalo/Lgal.
As an example, we show the fit to NGC4631 in Figure 11. The thermal emission from extraplanar dust provides a more
The top panels show the outcome of the MCRT simulation. In straightforward measurement of Mdust because the flux is
the upper left panel, noise comparable to that in the uvw2 image directly related to the column density. Here we use the 160 μm
was added to show how far the simulated halo could be images (as opposed to the other FIR bands) because 160 μm is
detected in the UVOT images (3 and 6σ contours are shown). close to the peak of the FIR SED for cool dust and because
24
there is more coverage of our sample with Herschel/PACS at If the dust composition changes with height (for example, if
this wavelength. the fraction of silicate grains increases), the reflected spectrum
In most cases, the 160 μm emission follows the UV halo in could become bluer given a constant incident spectrum. Since
the sense that they are bright in the same regions. However, the the optical depth through the halo is small (as is clear from
FIR emission is often not as extended vertically or radially as observing face-on galaxies), this is a good approximation. On
the UV halo (the exceptions being NGC 891 and NGC 4096). the other hand, given that the nuclear starburst is obscured
This could be because of the much lower UV background or through the disk but (presumably) not along the minor axis, the
because most of the FIR maps are shallow snapshots intended incident spectrum into the halo may differ across the disk,
to measure emission from the disk, not the halo, as part of leading to a projected change in the color. In this case,
Herschel surveys (such as the Herschel Reference Survey; sightlines vertically above the starburst region would reflect a
Boselli et al. 2010). bluer incident spectrum than the rest of the disk. This behavior
The dust mass can be obtained from the flux, provided one can be generalized to H II regions in the disk.
knows the temperature and grain emissivity law β. The number We cannot rule out either possibility, but in M82 the data
of grains is obtained from the flux and these quantities: suggest that the color change in height results from a change in
L n = Ndust 4pa2Qn (a) pBn (T ) , (2 ) the dust. Let us assume that the halo dust does not change.
Then, for a partially obscured nuclear starburst viewed edge-
β
where Qν(a)≈(2π a ν/c) is the dust emissivity and a the on, in which light can escape along the minor axis, we expect
grain size. The mass implied is the eRN to form bright lobes above the starburst and near the
disk. These will be visible against the rest of the eRN. The
4 lobes are bright near the disk because both the incident flux and
Mdust = Ndust pa 3rdust , (3 )
3 column density are higher there. Lobes of bluer color than the
rest of the eRN will also occur, with a small height offset
where ρdust is the intrinsic density of the material. We assume
between the bluest and brightest region. This offset occurs
that a=0.1 μm and that ρdust=2 g cm−3. There is actually a because the contribution to the eRN from the rest of the disk
distribution of grain sizes that can be constrained with a dust (with a redder incident spectrum) is also greatest near the disk.
model, which we defer to PaperIII. However, we do not know Above the blue lobes, the halo becomes redder as the incident
the temperature or β. T is usually measured by fitting a flux from the starburst declines and the reflected flux from this
modified blackbody to the far-infrared and sub-mm SED, but region competes with the flux from the remainder of the disk.
few galaxies in the sample have sufficient data to measure T in We confirmed these features through MCRT models of
the halo. partially obscured starbursts, as shown in Figure 12. We tried
For an order-of-magnitude estimate for comparison with the two models: a model with the same setup as described above
MCRT models, we assume that β=1.5 (in the literature it except with 90% of the emissivity clustered in the inner 1 kpc,
varies from 1 to 2 in nearby galaxies; e.g., Bendo et al. 2003; and a variation of this model with a “wind cone” of denser
Casey 2012), and use the temperature of the cold dust measured material. To examine the color we assumed that the starburst
in the disk for each galaxy. If the dust is embedded in neutral region has a Starburst2 spectrum and the remainder of the disk
gas, its temperature will likely decline with height, so these has an Sc spectrum from the Kinney–Calzetti atlas (Kinney
Mdust values will be lower limits. et al. 1996). We do not simulate a full spectrum, but rather
We measured 160 μm fluxes above a projected height of determined the proper relative luminosities for the FUV and
2 kpc for NGC891, NGC4631, and NGC5907 from Herschel NUV bands. The cone has an opening angle of 50°, a width of
maps (NGC 5775 does not have Herschel data, and the angular 0.8 kpc, and a density that declines exponentially with height
resolution of Spitzer/MIPS is too poor to isolate halo flux). but is uniform at each height within the cone region. The
These fluxes are 11.1±0.2 Jy for NGC891, 5.4±0.3 Jy for vertical profile is Ae-b z , where A is 10% of the central density
NGC4631, and 3.2±0.4 Jy for NGC5907. For temperatures constrained by the HI data, and the scale height b=1 kpc.
we use 23 K for NGC891 (Hughes et al. 2014), 22 K for These parameters are motivated by the appearance of limb-
NGC4631 (Meléndez et al. 2015), and 18 K for NGC5907 brightened winds in Hα, but are not fits to the data. We tried
(Dumke et al. 1997). Under these assumptions, the dust masses one additional variation in which the interior of the cone has
are Mdust=3.2×106, 0.7×106, and 6.1×106 for zero density (i.e., it lacks dust).
NGC891, NGC4631, and NGC5907, respectively (Table 8). In the case without the wind cone the main feature is the
These values are very sensitive to β; if β=2, the masses are appearance of the lobes described above, whereas the
about ten times larger. This highlights the need for deeper FIR remainder of the eRN looks like that for a normal disk galaxy.
data to constrain β in the extraplanar dust. Considering the When the cone is present, it is brighter than the rest of the halo
large uncertainties, the masses are consistent with the MCRT and its center is bluer than average, while the projected sides
values but do not provide useful constraints. are relatively red. However, the presence of a cone does not
change the presence of lobes in the FUV–NUV color map even
6.4. Dust Spatial Variation if the interior of the cone is dust-free because the light is
Some UV halos vary in FUV–NUV color with galactocentric scattered by the dense cone walls.
radius (Figures 6 and 7) or height (Figure 10). If the UV halos M82 has bright lobes near the disk and there is indeed an
are eRN, these differences arise either through changes in the offset between these lobes and the bluer region above, but there
dust or the incident spectrum, although both effects may be are no corresponding lobes in the FUV–NUV map and the
present. M82 provides a good example of the general behavior color instead continues to become bluer with height and covers
seen in starbursts, namely that the eRN is bluer in the wind an increasingly large fraction of the projected area (Figure 12).
region (Figure 7) and becomes bluer with height (Figure 10). These features cannot be explained by the simple geometric
25
Figure 12. Top left: combined UV intensity map for M82. Top right: FUV–NUV color map for M82. Blue or purple colors indicate where the halo is bluer, while
redder colors indicate larger FUV–NUV (in magnitudes). Bottom left: an MCRT simulation of an obscured starburst with a wind cone and a redder disk. The brightest
region of contours from the blue light are shown. A horizontal box indicates the galaxy disk. Bottom right: a synthetic FUV–NUV color map showing the halo colors
with the contours from left superimposed. We expect blue “lobes,” which are not seen in M82.
model above. The optical depth may be higher than we assume, larger heights. Since some of the FUV emission comes from
which would suppress lobes in the color map, but the halo at extraplanar star formation in these few cases, we speculate that
larger heights would also be redder as the incident spectrum the difference is not explained by a change in the dust with
would change substantially through the halo. Some of the FUV height.
light may come from line emission, but the same morphology is If the FUV–NUV color change with height around starbursts
seen in the very deep uvw2 image. The simplest explanation of results from a change in the dust, two plausible explanations
the color map is a change in the dust with height. This could are that galactic winds eject a different type of dust and carry it
either be a physical change in a single dust component with to larger heights than the processes in normal galaxies, or that
height or an increasing contribution from one component winds and a comparatively intense radiation field alters the pre-
of dust. existing dust at large heights. In normal galaxies, the dust may
In other cases, such as NGC253 or NGC3079, there are circulate through a galactic fountain (Bregman 1980), which is
blue regions in the halo near the bases of the winds that may consistent with the measured UV scale heights of several kpc.
indicate a different light source rather than a difference in dust
On larger scales, we expect the dust to change from Milky
(Figure 7). However, in both of these galaxies and in the
Way-like dust in the disk to SMC-like dust by a radius of
composite starburst measurement, FUV–NUV color increases
20 kpc (Ménard et al. 2010), which (for a given radiation field)
within a few kpc of the disk and then decreases with height
(Figure 10). This suggests that these filamentary structures exist will lead to smaller FUV–NUV color (however, see Smith
alongside an overall change in the dust with height. In contrast, et al. 2016, who find that most “extragalactic” dust can be
most of the normal galaxies do not have significant FUV–NUV explained by an extended disk). In the composite (Figure 10)
structure in Figure 6 and the composite FUV–NUV color does we do not detect the transition through FUV–NUV color alone,
not appear to change with increasing height. This suggests that but most of the galaxies do not have reliable FUV–NUV colors
the dust has similar properties between 2 and 12 kpc. However, above 10 kpc; the values there for normal galaxies come from 2
NGC891 and NGC5907 have high S/N data and show a to 4 galaxies, depending on the height. A clean test for this
modest decline with height above 5 kpc. Deeper data for more transition requires deeper data for more individual galaxies. If
galaxies are necessary to determine whether the color is the dust does not change around normal galaxies up to heights
constant with height. Finally, in Figure 10 the stripped galaxies difficult for a galactic fountain to reach, one possibility is that
appear bluer than starburst galaxies at low heights and redder at starburst episodes provide most of the halo dust.
26
Table 9 diffuse dust mass between 0.5 and 2 kpc in our MCRT
Swift Mrk 501 Observations models, which is the region where the filaments are seen.
ObsIDs While they argue that it would be challenging for a galactic
fountain (Bregman 1980) or radiation pressure (Ferrara
uvm2 uvw1 uvw2
et al. 1991) to lift enough dust mass to account for the
30793006 30793001 30793006 extinction features they see, either mechanism could produce
30793007 30793002 30793007 the smaller amount of diffuse dust inferred at large heights.
30793008 30793003 30793008 This suggests that the diffuse medium above 2 kpc is a
30793009 30793004 30793009
separate component.
30793010 30793006 30793010
30793011 30793007 30793011
M82 has one of the first UV halos reported (Hoopes
30793012 30793008 30793012 et al. 2005), since its extraplanar diffuse UV light is obvious
30793013 30793009 30793013 from the raw image. It is also the best-studied UV halo (Coker
30793014 30793010 30793014 et al. 2013; Hutton et al. 2014), and the dust content of the
30793015 30793011 30793015 wind has also been investigated by several authors (e.g.,
30793016 30793012 30793016 Engelbracht et al. 2006; Roussel et al. 2010; Yoshida
et al. 2011). The consensus that the UV halo in M82 is an
Note. Swift observations of Mrk501 used to construct the UVOT PSF models eRN supports the identification of other UV halos as eRN. The
used in this paper. FIR dust emission is seen to about the same or larger distances
(This table is available in its entirety in machine-readable form.) than in the UV, implying a mass of about 106 Me (Roussel
et al. 2010). It is more difficult to construct an accurate MCRT
model for M82 than other disk galaxies because of its irregular
6.5. M82 and NGC 891 H I morphology, but using the same method as described above
+3
NGC891 and M82 both have UV halos that have been we obtain a value of Mdust = 2- 1 ´ 10 M, which is consistent
6
previously studied, and here we compare our maps and results with the FIR measurement. M82 also allows us to test our
to those in the literature. NGC891 is a normal Milky Way de-reddening procedure: our reported FUV Lgal=3×
analog with an elevated star formation rate near 4 Me yr−1, 1042 erg s−1, whereas Coker et al. (2013) estimate that the
(Popescu et al. 2004) and a bright X-ray halo (Bregman & UV luminosity seen by the halo is Lgal=1–6×1042 erg s−1.
Pildis 1994), whereas M82 is the archetypal superwind galaxy. We suggest that the best simple explanation for the the
The UV halo around NGC891 was modeled by Seon et al. FUV–NUV color change with height (in M82 and starburst
(2014), who found that the UV light is consistent with a scale halos generally) is a change in the dust. We note that Roussel
height between 1.2–2 kpc and a dust mass above a projected et al. (2010) show that there is little change in the 250 μm/
height of 2 kpc of 3%–5% of the total. Hughes et al. (2014) find 350 μm ratio in the region with the UV halo, but if the change
a dust mass in the disk of Mdust=8.5×107 Me, so the halo is primarily one of composition rather than size then the FIR
dust mass from the MCRT models in Seon et al. (2014) is bands will not be sensitive to it. For example, Nozawa &
between 2.5×106 and 4.3×106 Me. This is consistent with Fukugita (2013) successfully fit a two-component (graphite and
the MCRT scale height and mass we measured: silicate) dust model to extinction along different Milky Way
h=1.5±0.2 kpc and Mdust=(6 ± 3)×106 Me (for Milky and SMC sight lines by changing the mixture but fixing the size
Way dust). Seon et al. (2014) fit a profile along the projected distribution using the same law found by Hutton et al. (2014).
Likewise, Hutton et al. (2014) found, using shallow Swift data,
minor axis, incorporating the core of the PSF but not the highly
that the dust is consistent with a single size distribution. They
extended wings.
focused on color–color plots where a change in composition is
In HKB14, where we did not account for the extended PSF
degenerate with a change in the incident spectrum, so neither of
contamination of the halo, the UV intensity falls off more
these studies presents strong evidence against a change in the
slowly with height on the side with the large H I filament,
dust type with height. However, we emphasize that we have not
which also occurs in the X-rays and the 160 μm image. The conclusively shown that the dust composition must change.
combined, PSF-corrected UV map shows that the part of the
UV halo clearly associated with NGC891 is brighter on that
side (Figure 3), but the large-scale diffuse UV light is too 7. SUMMARY AND CONCLUSIONS
extensive to be scattered light from dust around NGC891 and
may be Galactic cirrus. The increased UV intensity near the HI We have described the morphology of UV halos around
filament may arise from dust within the filament material nearby, edge-on galaxies with and without superwinds, and
adding to the total dust column. here summarize our findings:
Howk & Savage (1997) examined extraplanar dust around 1. UV halos are astrophysical and can be separated from
NGC891 using extinction (at a lower height than the UV galactic light scattered into the PSF wings (Airy patterns).
halo), and found filamentary structures below 2 kpc that have They are broadband phenomena and truly diffuse, with a
a mass of at least a few ×107 Me, or about 10%–50% of the flux of 1%–20% of the apparent galaxy luminosity (a few
total (Hughes et al. 2014). The resolution of their images is percent or less of the de-reddened values). They are
much higher than the UV images, so we cannot rule out the visible to beyond 10 kpc above the midplane.
possibility that the eRN is also highly structured. However, 2. Around normal galaxies, UV halos tend to have a thick-
the mass in the scattering component (estimated through disk morphology, but they differ in their radial concen-
MCRT or 160 μm flux) is less than half than that inferred by tration and vertical prominence. In superwind galaxies,
Howk & Savage (1997). This is also true if we measure the the UV halos are visible around the whole galaxy and
27
Table 10
GALEX and Swift Observations in this work
Galaxy ObsIDs
FUV NUV uvw2 uvm2 uvw1
M82 GI1_071001_M81 GI1_071001_M81 31201001 31201001 31201001
32503004 31201002 32503002
32503016 32503003 32503015
32503020 32503007 32503018
32503028 32503008 32503022
32503032 32503011 32503025
32503036 32503017 32503026
32503040 32503021 32503034
32503046 32503024 32503043
32503057 32503029 32503045
Note. Swift and GALEX observations used in this paper.

(This table is available in its entirety in machine-readable form.)
contain filamentary structures seen at other wavelengths chemical composition, temperature, etc.) tell us about the role
and associated with the winds. Galaxies that are being of stellar (or possibly AGN) feedback in the history of the
stripped of their ISM by ram pressure have asymmetric galaxy and the pollution of the circumgalactic medium with
UV halos and some extraplanar star formation. metals. However, much work remains: a larger, less biased
3. Among a variety of galactic parameters, we found that the sample of UV halo properties is needed, at the cost of poorer
UV halo luminosity is only correlated with the galaxy resolution. This will be addressed in PaperII. In PaperIII, we
luminosity and the SFR. will constrain dust properties in a simple model using the UV
4. The structure, broadband visibility, and strong correlation SED. Beyond this, we need to understand the dust outflow
of Lhalo with Lgal for UV halos lead us to conclude that mechanism (the roles of radiation pressure and hydrodynamic
they are eRN, which supports prior arguments made in entrainment), connect the UV SED to the FIR SED, and
the literature. determine if (or how) the dust within 10 kpc of the galaxy
5. The frequency of UV halos (100% in our UV-selected connects to a larger circumgalactic component.
sample of highly inclined galaxies within 25 Mpc)
indicates that eRN are ubiquitous, but not as extensive We thank the anonymous referee for a helpful report that
as reported in HKB14. improved the quality of this paper. This research has made use
6. The dust mass of the diffuse component above 2 kpc, of the NASA/IPAC Extragalactic Database (NED) which is
inferred from MCRT models that are constrained by the operated by the Jet Propulsion Laboratory, California Institute
measured Lhalo/Lgal and 21 cm maps, is a few percent of of Technology, under contract with the National Aeronautics
the dust mass in the disk, and perhaps 10% of the dust and Space Administration. This research has made use of data
mass seen in filamentary extraplanar structures at lower and software provided by the High Energy Astrophysics
heights. Deeper FIR observations are needed to constrain
Science Archive Research Center (HEASARC), which is a
β in the halo dust for a direct measurement.
service of the Astrophysics Science Division at NASA/GSFC
7. There is tentative evidence for a change in the dust
and the High Energy Astrophysics Division of the Smithsonian
properties with height in starburst galaxies, but normal
Astrophysical Observatory. Some of the data presented in this
galaxies are consistent with constant dust properties
between 2 and 12 kpc. paper were obtained from the Mikulski Archive for Space
Telescopes (MAST). STScI is operated by the Association of
We expect virtually every star-forming galaxy to form an eRN, Universities for Research in Astronomy, Inc., under NASA
but their detectability in a given sample is a strong function of contract NAS5-26555. Support for MAST for non-HST data is
the UV luminosity of the galaxy and the inclination. Beyond provided by the NASA Office of Space Science via grant
about 25 Mpc it becomes difficult to isolate filamentary NNX09AF08G and by other grants and contracts. We
structure with GALEX or Swift, but within this limit the acknowledge the usage of the HyperLeda database (http://
UV halos around normal galaxies appear smooth, and there leda.univ-lyon1.fr). EH-K and JC gratefully acknowledge
are no clear features in the FUV–NUV color maps. One support from NASA grant NNH13ZDA001N-SWIFT.
consequence is that the height profile used to constrain MCRT Facilities: GALEX, Swift.
models (Seon et al. 2014; Shinn & Seon 2015) provides
a reasonable estimate of the dust mass, assuming a thick
disk morphology. UV observations of more distant starburst or APPENDIX
wind galaxies may not discern the wind structure, but an
unusually extensive or blue UV halo may indicate the presence
of a wind. Tables 9 and 10 contain the Swift and GALEX observation
Our results confirm the presence of diffuse dust around IDs for the Mrk501 PSF measurements and the galaxies used
galaxies as a general phenomenon. The dust is a reliable tracer in this work, respectively. The GALEX PSF models were based
of material that, at one point, originated in a galaxy disk, so the on data sets for 3C273 (obsID GI4_012003_3C273) and
amount and physical properties (grain size distribution, PKS2155–304 (obsID PKS2155m304).
28
REFERENCES Koopmann, R. A., Kenney, J. D. P., & Young, J. 2001, ApJS, 135, 125
Kriek, M., & Conroy, C. 2013, ApJL, 775, L16
Abramson, A., Kenney, J. D. P., Crowl, H. H., et al. 2011, AJ, 141, 164 Lee, S. W., & Irwin, J. A. 1997, ApJ, 490, 247
Allaert, F., Gentile, G., Baes, M., et al. 2015, A&A, 582, A18 Lehnert, M. D., & Heckman, T. M. 1996, ApJ, 462, 651
Aniano, G., Draine, B. T., Gordon, K. D., & Sandstrom, K. 2011, PASP, Li, J.-T., & Wang, Q. D. 2013, MNRAS, 428, 2085
123, 1218 Lucero, D. M., Carignan, C., Elson, E. C., et al. 2015, MNRAS, 450, 3935
Baes, M., & Viaene, S. 2016, A&A, 587, A86 Makarov, D., Prugniel, P., Terekhova, N., Courtois, H., & Vauglin, I. 2014,
Bell, E. F., & de Jong, R. S. 2001, ApJ, 550, 212 A&A, 570, A13
Bendo, G. J., Joseph, R. D., Wells, M., et al. 2003, AJ, 125, 2361 Martin, D. C., Fanson, J., Schiminovich, D., et al. 2005, ApJL, 619, L1
Benjamini, Y., & Hochberg, Y. 1995, J. R. Stat. Soc. Ser. B, 57, 289 McCormick, A., Veilleux, S., & Rupke, D. S. N. 2013, ApJ, 774, 126
Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393 Meléndez, M., Veilleux, S., Martin, C., et al. 2015, ApJ, 804, 46
Boselli, A., Eales, S., Cortese, L., et al. 2010, PASP, 122, 261 Ménard, B., Scranton, R., Fukugita, M., & Richards, G. 2010, MNRAS,
Bottema, R., van der Kruit, P. C., & Freeman, K. C. 1987, A&A, 178, 77 405, 1025
Bregman, J. N. 1980, ApJ, 236, 577 Nozawa, T., & Fukugita, M. 2013, ApJ, 770, 27
Bregman, J. N., & Pildis, R. A. 1994, ApJ, 420, 570 Oke, J. B., & Gunn, J. E. 1983, ApJ, 266, 713
Buat, V., Donas, J., Milliard, B., & Xu, C. 1999, A&A, 352, 371 Oosterloo, T., Fraternali, F., & Sancisi, R. 2007, AJ, 134, 1019
Calzetti, D. 2001, PASP, 113, 1449 Popescu, C. C., Tuffs, R. J., Kylafis, N. D., & Madore, B. F. 2004, A&A,
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682 414, 45
Casey, C. M. 2012, MNRAS, 425, 3094 Rand, R. J. 1996, ApJ, 462, 712
Chung, A., van Gorkom, J. H., Kenney, J. D. P., Crowl, H., & Vollmer, B. Rice, W., Lonsdale, C. J., Soifer, B. T., et al. 1988, ApJS, 68, 91
2009, AJ, 138, 1741 Roming, P. W. A., Kennedy, T. E., Mason, K. O., et al. 2005, SSRv, 120,
Coker, C. T., Thompson, T. A., & Martini, P. 2013, ApJ, 778, 79 95
Collins, J. A., Rand, R. J., Duric, N., & Walterbos, R. A. M. 2000, ApJ, Rossa, J., & Dettmar, R.-J. 2003a, A&A, 406, 493
536, 645 Rossa, J., & Dettmar, R.-J. 2003b, A&A, 406, 505
Condon, J. J. 1987, ApJS, 65, 485 Roussel, H., Wilson, C. D., Vigroux, L., et al. 2010, A&A, 518, L66
de Vaucouleurs, G., de Vaucouleurs, A., Corwin, H. G., Jr., et al. 1991, Third Sandin, C. 2015, A&A, 577, A106
Reference Catalogue of Bright Galaxies, ed. N. G. Roman et al. (New York: Schlafly, E. F., & Finkbeiner, D. P. 2011, ApJ, 737, 103
Springer) Seon, K.-i., Witt, A. N., Shinn, J.-h., & Kim, I.-j. 2014, ApJL, 785, L18
Dumke, M., Braine, J., Krause, M., et al. 1997, A&A, 325, 124 Shinn, J.-H., & Seon, K.-I. 2015, ApJ, 815, 133
Engelbracht, C. W., Kundurthy, P., Gordon, K. D., et al. 2006, ApJL, Smith, M. W. L., Eales, S. A., De Looze, I., et al. 2016, arXiv:1607.01020
642, L127 Strickland, D. K., Heckman, T. M., Colbert, E. J. M., Hoopes, C. G., &
Ferrara, A., Bianchi, S., Dettmar, R.-J., & Giovanardi, C. 1996, ApJL, 467, L69 Weaver, K. A. 2004, ApJS, 151, 193
Ferrara, A., Ferrini, F., Barsella, B., & Franco, J. 1991, ApJ, 381, 137 Tüllmann, R., Pietsch, W., Rossa, J., Breitschwerdt, D., & Dettmar, R.-J. 2006,
Haynes, M. P., Giovanelli, R., & Roberts, M. S. 1979, ApJ, 229, 83 A&A, 448, 43
Hodges-Kluck, E., & Bregman, J. N. 2014, ApJ, 789, 131 Vollmer, B., Nehlig, F., & Ibata, R. 2016, A&A, 586, A98
Hoopes, C. G., Heckman, T. M., Strickland, D. K., et al. 2005, ApJL, 619, L99 Walter, F., Dahlem, M., & Lisenfeld, U. 2004, ApJ, 606, 258
Howk, J. C., & Savage, B. D. 1997, AJ, 114, 2463 Weingartner, J. C., & Draine, B. T. 2001, ApJ, 548, 296
Howk, J. C., & Savage, B. D. 1999, AJ, 117, 2077 Weliachew, L., Sancisi, R., & Guelin, M. 1978, A&A, 65, 37
Hughes, T. M., Baes, M., Fritz, J., et al. 2014, A&A, 565, A4 Westmeier, T., Koribalski, B. S., & Braun, R. 2013, MNRAS, 434, 3511
Hutton, S., Ferreras, I., Wu, K., et al. 2014, MNRAS, 440, 150 Whitney, B. A. 2011, BASI, 39, 101
Irwin, J. A. 1994, ApJ, 429, 618 Wiegert, T., Irwin, J., Miskolczi, A., et al. 2015, AJ, 150, 81
Irwin, J. A., Seaquist, E. R., Taylor, A. R., & Duric, N. 1987, ApJL, 313, L91 Wood, K., Smith, D., Whitney, B., et al. 2001, ApJ, 561, 299
Kenney, J. D. P., & Koopmann, R. A. 1999, AJ, 117, 181 Yim, K., Wong, T., Xue, R., et al. 2014, AJ, 148, 127
Kenney, J. D. P., & Koopmann, R. A. 2001, in ASP Conf. Ser. 240, Gas and Yoshida, M., Kawabata, K., & Ohyama, Y. 2011, PASJ, 63, 493
Galaxy Evolution, ed. J. E. Hibbard, M. Rupen, & J. H. van Gorkom (San Young, J. S., Allen, L., Kenney, J. D. P., Lesser, A., & Rownd, B. 1996, AJ,
Francisco, CA: ASP), 577 112, 1903
Kennicutt, R. C., Jr. 1998, ApJ, 498, 541 Yun, M. S., Ho, P. T. P., & Lo, K. Y. 1994, Natur, 372, 530
Kennicutt, R. C., Jr., Lee, J. C., Funes, J. G., et al. 2008, ApJS, 178, 247 Zschaechner, L. K., Rand, R. J., Heald, G. H., Gentile, G., & Józsa, G. 2012,
Kinney, A. L., Calzetti, D., Bohlin, R. C., et al. 1996, ApJ, 467, 38 ApJ, 760, 37
29

Hodges-Kluck 2016 ApJ 833 58

Uploaded by

Copyright:

Available Formats

Hodges-Kluck 2016 ApJ 833 58

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hodges-Kluck 2016 ApJ 833 58

Uploaded by

Copyright:

Available Formats

The Astrophysical Journal, 833:58 (29pp), 2016 December 10 doi:10.

ULTRAVIOLET HALOS AROUND SPIRAL GALAXIES. I. MORPHOLOGY

Name Type T i MB d vrot E (B - V ) B−V mK M* LHα (1040 SFR(IR)

Name Exposure Time (s) Sensitivity (AB mag) Hα Radio 21 cm

NGC 0253 3281 15043 L L L 25.0 26.1 L L L L 5 13

NGC 0055 29347 30691 L L L 25.5 26.5 L L L L L 20

Filter Exponent Filter Exponent

Name FUV NUV uvw2 uvm2 uvw1 uvw2 Factor

NGC 0253 105.20±0.01 200.78±0.02 L L L 0.93

NGC 0055 294.56±0.01 441.60±0.03 L L L 0.93

Halo Flux (mJy) PSF-Wing Contamination (mJy) uvw2 Factor

NGC 0253 5.43±0.09 17.2±0.2 L L L 4.89 0.60 L L L 0.93

Extraplanar Diffuse? Lhalo Lgal Lhalo/Lgal

NGC 0253 1.26±0.02 2.2±0.1 38 Y Y Y Y 67.6 3.6 L 6.9 L 1448.2 L 0.005 L

Hodges-Kluck, Cafmeyer, & Bregman

Thick Disk Milky Way Dust SMC Dust

Note. Swift and GALEX observations used in this paper.

You might also like