© The Authors 2022

1. Introduction

It has been known for many years that spiral galaxies need an ongoing gas supply in order to explain several observational properties. Among others, the observed star formation rates (SFRs) and histories (Daddi et al. 2007; Bothwell et al. 2011), the chemical composition of the Galactic disk (Chiosi 1980) and large-scale disturbances of galactic disks, for example, lopsidedness or warps (van Eymeren et al. 2011), are readily explained by a continuous accretion of gas. Further observational evidence for the continuous accretion of cold material, beyond the direct surroundings of the observed disks, comes from the fact that H I-excess galaxies reside in H I-overdense regions (Wang et al. 2015b) and that the H I content in galaxies appears similar to their neighbors, the so-called H I conformity (Wang et al. 2015a). Furthermore, on even larger scales, the decline of the Universal H I mass density as a function of time is almost negligible compared to the build-up of stars over the same period (Putman 2017). This can only be explained with material from the intergalactic medium (IGM) and galaxy halos condensing onto the H I disks (Putman et al. 2012).

The neutral gas accretion rates observed to date have been unable to match the levels required to explain these aforementioned observational properties. In the local Universe, a global H I accretion rate of 0.2 M_⊙ yr⁻¹ was estimated by Sancisi et al. (2008) based on observational findings. This accretion rate includes both the accretion of pure gas clouds and gas-rich dwarf galaxies and is based on the literature of single-target studies. Di Teodoro & Fraternali (2014) refined this estimate by systematically searching the Westerbork H I Survey of Irregular and Spiral sample (WHISP, van der Hulst 2002) for companion galaxies that could be accreted. The upper limit for gas accretion derived in this study (0.28 M_⊙ yr⁻¹) was several times lower than the average SFR in the sample (1.29 M_⊙ yr⁻¹), showing that minor mergers cannot supply the required amounts of neutral hydrogen in this simple comparison.

Unfortunately, in only a handful of objects have signs of accretion been studied in detail. It has been known since the early 1960s that the Milky Way halo contains significant amounts of neutral hydrogen in the form of high-velocity clouds (HVCs; Muller et al. 1963; Wakker & van Woerden 1997) and the potentially more distant ultra-compact HVCs (Adams et al. 2013). Only in the past decade or so have analogous features been observed around other nearby spiral galaxies as well, for instance, in M 31 (Westmeier et al. 2008), NGC 2403 (Fraternali et al. 2002), NGC 891 (Oosterloo et al. 2007) and NGC 6946 (Boomsma et al. 2008), which indicates that these large and distinct H I features are not uncommon. However, only a few large-scale clouds or filaments are typically seen in the surroundings of a single galaxy, and the bulk of them resides close to the disk. Thus, from these detailed studies and others, a picture has emerged in which only a fraction of this gas originates from outside the galaxy disks (Fraternali & Binney 2008; Fraternali et al. 2015).

From the theoretical perspective, continuous accretion of gas onto the galaxy disks is also predicted, but it is unclear whether the majority of this accretion occurs in a cold or in a hot phase. Especially for lower redshifts (z < 1), the debate is ongoing whether in lower-mass halos (M_vir < 10¹¹ M_⊙) the dominant mode is the accretion of gas that never passed through a hot phase (i.e., cold-mode accretion; Nelson et al. 2013; Huang et al. 2019) or gas that did (i.e., hot-mode accretion; Nelson et al. 2015; Schaller et al. 2015). The answer crucially depends on the type of feedback included in the simulations. Additional observational evidence can therefore help constrain the simulations. In this it is important to note that some authors defined the cold phase as gas that is never heated above 2.5 × 10⁵ K (Huang et al. 2019), whereas others defined it as gas that has never exceeded the virial temperature of the halo it is being accreted onto (Nelson et al. 2013). Both definitions allow the cold phase to include significant fractions of ionized hydrogen, that is, gas with temperatures > 10⁴ K, which even further complicates a comparison of H I observations to these predictions. Additionally, simulations have not yet reached convergence, that is, various properties of the cold gas depend significantly on the adopted resolution, especially when considering the circumgalactic medium (CGM; van de Voort et al. 2019).

In order to quantify the typical neutral gas accretion rate in a range of spiral galaxies, we have embarked on the Westerbork Hydrogen Accretion in LOcal GAlaxieS (HALOGAS) survey to investigate general characteristics of gaseous halos and accretion in a significant sample (Heald et al. 2011b). For this purpose, we obtained deep (10 × 12 h) H I observations using the Westerbork Synthesis Radio Telescope (WSRT), reaching typical column density sensitivity levels of ∼10¹⁸ cm⁻² in ∼ 4.12 km s⁻¹ channels and a spatial resolution of 20″–30″ (see Heald et al. 2011b; Heald & HALOGAS Team, in prep., for further details). Here, we present the results of a systematic search for distinct intergalactic clouds and streams in the vicinity of the sample galaxies using the SOurce Finder Application (SOF IA; Serra et al. 2015). Using SOF IA has distinct advantages over a visual inspection because it is unbiased, quantifiable, and reproducible. As the HALOGAS sample includes a broad range of galaxy properties, our findings allow us to estimate global neutral gas accretion parameters.

This paper is structured as follows. In Sect. 2 we describe the method that we used to analyze HALOGAS cubes in order to find cloud candidates in a robust manner. In Sect. 3 we present the tabulated result of our search and present ancillary data for the targets and how they relate to our search result. In Sect. 4 we present the results derived from the full sample, and in Sect. 5 we discuss our findings. Finally, in Sect. 6 we summarize and conclude our analysis.

2. Method

In order to find gas clouds around the galaxies in the HALOGAS sample, we made use of the H I source finder application SOF IA (v1.3.1; Serra et al. 2015; Wang et al. 2015b; Westmeier et al. 2021; Józsa et al. 2022). SOF IA is an extended package that has many different settings that can be fine-tuned to optimize the detection of different types of sources.

2.1. Artificial source-finding test

First we investigated the optimal SOF IA parameter settings for finding small unresolved clouds by searching a data cube that contained 250 artificial point sources. This also provided us with a limiting flux threshold of our final source-finding method.

To create a data cube that resembles the HALOGAS observations as closely as possible, we first created an empty model cube with the same size, coordinates, and spectral resolution as the actual full-bandwidth data cube of the HALOGAS observations of NGC 3198 (Heald et al. 2011b; Gentile et al. 2013). NGC 3198 was chosen as its data cube is representative for most of the data cubes in the survey. It has the HALOGAS standard pixel size of 4″ × 4″ and channel width. This cube is populated with randomly distributed model H I sources, each of which is considered to be unresolved, that is, a point source centered on a single pixel convolved with the point spread function (PSF; i.e., the synthesized beam). Although the point sources were injected randomly, care was taken that they were separated by at least four PSF full widths at half maximum (FWHM; see Table 1) and five channels. This was done to prevent overlapping sources in the final artificial cube, which might confuse our detection statistics.

In this test, we focused on point sources because the main aim of this paper is to search the HALOGAS survey for clouds without optical counterparts. These clouds are expected to be of a physical size that is not resolved by the WSRT synthesized beam in most cases. Additionally, experience taught us that in SOF IA, even when settings are not optimized toward extended sources, it is unlikely that an extended source is missed completely when settings are optimized for point sources.

The H I mass of each source was chosen randomly in the range 4 < log(M_H I) < 5.7 assuming a uniform distribution in log space. The systemic velocity was randomly selected in the range 100–1472 km s⁻¹, with the upper limit corresponding to the maximum velocity in the cube. This velocity was converted into a distance by assuming a pure Hubble-Lemaître flow with H₀ = 70 km s⁻¹, and subsequently, the flux was calculated. Along the spectral axis, the artificial sources had a Gaussian distribution with an FWHM in the range 15–30 km s⁻¹ to represent the internal dispersion and some slight internal velocity structure in the clouds. If the initial mass of the artificial source was > 5 × 10⁴ M_⊙, this FWHM was increased with a factor $\sqrt{\frac{\mathrm{Mass}}{5\times {10}^4}}$ to resemble the larger velocity spread of more massive sources. These settings cause the peak flux densities to lie in the range of ∼0.003 − 30σ of the template cube, thus ensuring a range of sources from those that could not be missed to others that are impossible to detect. The sample of 250 artificial sources has the following mean values: mass = 1.2 × 10⁵ M_⊙, FWHM = 33.5 km s⁻¹, and an average peak flux density of 0.4 mJy beam⁻¹, which corresponds to 2.4σ_template.

The model cube, which is in the image domain, was then converted into the uv-plane with the task TCLEAN in the Common Astronomy Software Application (CASA; McMullin et al. 2007) for each observing run completed for NGC 3198. The model visibilities were doubled and then added to the XX correlation of the original calibrated and continuum-subtracted data, but not to the YY correlation. Assuming that the original data do not contain any polarized emission or polarization errors, a Stokes Q data cube created from these data retains all the noise properties and instrumental characteristics of the real observations, as well as our artificial sources, without including any real emission. This artificial Stokes Q cube was produced in exactly the same manner as the original Stokes I data cube of NGC 3198, that is, the inversion to the image domain and deconvolution was made with the same software and settings as in the real data. In both cubes, any corrections required due to the primary beam response were ignored. When converting back to the image domain, our point sources were effectively convolved with the dirty beam and thus were no longer contained in a single pixel.

We then ran SOF IA on this mock data cube with different settings. Of the tested parameters, the noise determination, merging maximum separation, merging minimum size, and reliability had small impact on the final results. The following parameters are critical, and hence we also list the investigated ranges:

• Detection threshold: The level below which the mask at each scale was clipped. We explored values between 3σ and 6σ.

• Kernels: The smoothing kernels used to search for emission in the data. We explored values in the spatial direction between 0 and 8 pixels and along the velocity axis from 0 to 16 channels.

• Kernel scale: The scale used to smooth the indicators that are used to determine the reliability of a detection. The full range from 0 to 1 was explored.

• Integrated signal-to-noise cut: In the reliability evaluation, it can transpire that positive sources are not real, but lie in an area of the noise distribution that is sparsely sampled. This could lead to false positives. Hence a more absolute cutoff on the integrated signal-to-noise ratio¹ was also applied. For our test set, we sampled a range of 15–35.

For more information about these parameters, we refer to the SOF IA website² and to Serra et al. (2015). We searched this grid of parameters by systematically varying one parameter in the search grid while keeping the others at the central value for the range. The best value was identified as the value that gave the highest ratio of real-to-false detections. After all optimal values were determined in this way, we ensured that the parameters could be optimized in this independent manner by once more running through the grid with all the optimal values set, albeit with smaller ranges. This did not lead to any change in the optimal parameters.

Table 2 gives the final parameters that resulted in the highest ratio of real-to-false detections. After source detection, the source masks were dilated until the integrated flux in the source increased by a factor lower than 0.02 compared to the previous mask, with a maximum of 3 pixels increase around the original detection.

With these settings, we retrieved 37 distinct model sources with SOF IA. Of these, one was a false detection. A careful inspection of this false detection showed that it corresponded to residual emission from NGC 3198, either due to small calibration errors in the data or to imperfect continuum subtraction (possibly related to polarized nonthermal emission). This leads to an important point: Our source finding is only as good as our data products, that is, bright artifacts will be detected by SOF IA. The 36 retrieved artificial sources all correspond to sources from the input catalog with an integrated flux higher than 21 mJy km s⁻¹, with a lowest spectral peak flux density of 0.43 mJy beam⁻¹, which corresponds to ∼2.7σ in a single channel in our mock cube.

One issue here is the small merging radius adopted in our settings. Experiments have shown that this is necessary to not miss small sources as the noise statistics for the reliability calculations change when sources are merged, making SOF IA less sensitive to the smaller sources. However, this means that real extended sources might break up into several sources. Visual inspection is required to identify and merge extended sources that are broken into segments by SOF IA.

Another important point to investigate in our source finding is how well the flux of a source can be retrieved after it is identified. Figure 1 shows the input integrated flux versus the retrieved integrated flux (circles) as well as the peak flux densities (stars). All sources lie closely to the line indicating identity, which means that when a source is found, SOF IA retrieves most of the flux associated with that source. At the lower end, the scatter indicates that noise does affect our detections. The input integrated flux of four model sources lies slightly above our nominal detection limit of 21 mJy km s⁻¹. These sources are not detected by SOF IA. A visual inspection shows that in the final cube, these sources are indeed much less significant than our faintest detections, most likely due to the distribution of noise peaks on top of these sources, that is, some lie on a positive noise peak, while others lie on a negative one. SOF IA was able to identify this reduced significance and exclude the sources as detections.

Fig. 1. Integrated flux (circles, mJy km s−1, right and bottom axes) and spectral peak flux densities (stars, mJy, left and top axes) from the input catalog vs. SOF IA-retrieved values for the detected sources. Black and gray are the sources detected by SOF IA, red is the false detection.

2.2. Searching the survey data cubes

The fact that SOF IA will also detect artifacts means that a final visual inspection of all SOF IA detections is required to determine whether they are artifacts. For this reason, we decided to assign to each source identified by SOF IA a numerical flag in the range 1–4 relating to our assessment of the reliability of the source. We used the following classification scheme:

• 1 - Real

• 2 - Likely real

• 3 - Likely not real

• 4 - Not real, artifact

The visual inspection was performed by one of us (PK) and was subsequently checked by several other authors. When a source was classified as real or likely real, the NASA Extragalactic Database (NED)³ and Simbad⁴ were queried to search for optical counterparts classified as galaxies. Around each source, we searched a cylinder with 500 km s⁻¹ depth and a radius defined by the major axis of the 3σ ellipse as fitted by SOF IA. The largest allowed radius for the cylinder was 2′ to avoid searching large catalogs of individual objects in well-resolved sources. When we found a source in this cylinder, we deemed it the optical counterpart to our detection. These optical counterparts are listed in the counterpart column of Table A.1. When no source was found, we inspected available optical images and queried the databases manually. Based on this, we estimated whether optical emission is present and, if so, whether it is a counterpart or a background source. These detections without spectroscopic confirmation are indicated with an asterisk in Table A.1.

To avoid complications by varying noise levels within a single cube, we ran our source finding on data cubes that had not been corrected for the primary beam response. Instead, we performed this correction on the integrated flux of the detected sources based on the distance of their central coordinates to the phase center. For this correction, we applied the standard correction for the WSRT, cos⁶(c × ν × r), with ν the frequency in Ghz, r the distance to the phase center in radians, and c a constant. For c we did not use the standard c = 68, but instead used c = 63 because it has been shown to better match the nonparameterized WSRT beam (Wang et al. 2015a). We list both the corrected and uncorrected flux in Table A.1.

Finally, to be able to run SOF IA on the real sample, we needed to apply some specific constraints to the source finding that differed from data cube to data cube. As the integrated S/N is based on a flux level in Jy beam⁻¹, the integrated flux S/N threshold needs to be corrected for the beam size. This was done by scaling this threshold with the ratio of the square root of the beam in the cube under investigation and the mock data cube. The thresholds used for each cube are shown in Table 1. Before running SOF IA on the real data, we first identified bad channels. These channels can occur at the edges of the cube or due to Milky Way emission. We identified them by calculating a mean rms level for the cube and then checking each channel against this mean. Channels in which the noise increased by more than 1% in three of the four corners were flagged. This typically resulted in a good identification of channels containing Milky Way emission and increased noise. However, in some cases, even this stringent criterion still missed channels with Milky Way emission. In these cases, the channel range being flagged was manually adjusted.

With these settings, we ran SOF IA on the 22 data cubes of the HALOGAS survey that cover their full 10 MHz bandwidth (∼2000 km s⁻¹), that is, not only the channels close to the target galaxy. These extended cubes are not publicly available due to their large size, but can be shared on reasonable request to the corresponding author. The normal HALOGAS cubes are available online⁵. The use of the extended cubes means that NGC 891 and NGC 2403 are not included in this analysis because only data cubes with a narrow range of velocities around the target are available.

3. Source-finding results

From the 22 Stokes I data cubes of the HALOGAS sample, we initially obtained 102 individual sources detected with SOF IA. After we combined artificially separated segments of single galaxies, a total of 80 sources were left. From these, we have 54 real sources, seven likely real detections, one likely not real detection, and 18 artifacts.

In order to not only judge the reliability of our detections, but also the type of detection, we also assigned a class identifier to each source during the visual inspection. These are defined as follows:

• Targets: The main body of the galaxies that were selected for the survey and are located in the center of the observations.

• Companions: These are detections with an optical counterpart separated from the target by more than one beam in all channels.

• Cloud candidates: The same as companions, but without an optical counterpart.

• Artifacts: Sources that can be identified as artifacts of the data reduction.

Of the sources that are real or likely real, 22 are the target galaxies, 25 are companions, and 14 are cloud candidates. Of the latter, 7 are classified real and 7 as likely real. The constructed catalog of the SOF IA-detected sources is presented in Table 3 for the cloud candidates and in Table A.1 for all other sources. The column entries are the following:

Col. (1) – Target galaxy.

Col. (2) – SOF IA-assigned source number within the target field.

Col. (3) – Right ascension of the flux-weighted centroid in J2000.

Col. (4) – Declination of the flux-weighted centroid in J2000.

Col. (5) – Class: T – target galaxy, C – companion, CC – cloud candidate, A – artifact.

Col. (6) – Peak flux density in mJy beam⁻².

Col. (7) – Integrated flux in Jy km s⁻¹.

Col. (8) – Integrated flux in Jy km s⁻¹, corrected for the primary beam.

Col. (9) – Line width at 20% of the peak flux density in km s⁻¹.

Col. (10) – Systemic velocity of the flux-weighted centroid, in km s⁻¹.

Col. (11) – Projected distance to main target in kpc.

Col. (12) – Flag: 1 – real; 2 – likely real; 3 – likely not real; 4 – not real, artifact.

Col. (13) – Identified optical counterpart (only applies to Table A.1).

In the appendix of this paper, we present various visualizations of the detections. In Appendix B.1 we present an overview image for every field with a detection in addition to the target galaxy and discuss some of the specifics of the individual detections and targets. In these images, the column density contours of all detections are overlaid on the R-band images of HALOSTARS (see Heald et al. 2011b) or R-band images taken at Kitt Peak National Observatory (KPNO; described in the atlas paper). If no image was available, the background is a DSS 2 R-band image. Additionally, we show in Figs. B.16–B.29 from left to right the H I column density map and intensity-weighted velocity field (also known as moment 0 and moment 1 maps), a position-velocity (PV) diagram along the morphological major axis as determined by SOF IA, and a line profile for all cloud candidates⁶. Finally, Figs. B.30–B.54 show moment maps of all other reliable sources except for the target galaxies. All data products relating to the targets will be presented in Heald & HALOGAS Team (in prep.) and are not reproduced here. These images also indicate the contour levels used in the overview images because they vary from source to source.

The overview images give an impression of the distribution of the retrieved sources around the targets. An excellent example of this is Fig. B.10. In this figure, all the different classes are represented. This image also shows that some detections are so close to the target galaxies that they can be considered as part of the galaxy halo (sources 3, 5, and 7), whereas others are farther removed and might not be associated with the target (sources 1, 4, 6, and 8). For the purpose of this paper, however, we only differentiate between detections based on the existence of a known optical counter part.

As we have optimized the source finding for point sources, at this point, we also verified that no extended sources were missed in the source finding by comparing the sources found in this analysis to those found in the HALOGAS galaxies that have undergone detailed individual studies (see Appendix B.1; Heald et al. 2011a; Zschaechner et al. 2011, 2012; Gentile et al. 2013; Kamphuis et al. 2013; de Blok et al. 2014; Vargas et al. 2017) and a visual inspection of the cubes by several of the authors.

In order to facilitate the discussion below, Table 4 shows an overview of properties for the target galaxies in the HALOGAS survey. The table does not list the quantities for NGC 672 and NGC 4631 because these systems are interacting, and hence they were excluded from the final analysis. The distances come from the original HALOGAS paper (Heald et al. 2011b).

The systemic velocity, W₂₀, R_max, and R_min are all as retrieved by our SOF IA run, where R_max and R_min are half the major and minor axis of the 3σ ellipse fitted by SOF IA. For consistency, we used the values from our SOF IA run. These values are slightly different from those published in the data release and the Atlas paper (Heald & HALOGAS Team, in prep.) because our cubes have a higher resolution and the SOF IA parameters are optimized for point sources. These differences are typically very small; the mean difference in W₂₀ is 1.6 km s⁻¹, for example. The stellar mass was calculated as the average from the methods and catalogs listed in the notes. The H I mass in solar masses was calculated from the corrected integrated flux as listed in Table A.1 and the distance (D [Mpc]) with M_H I = F_i,pbcorr ×2.36 × 10⁵ × D² (M_⊙). The SFRs were taken from the pilot paper (Heald et al. 2011b), and the depletion time (τ) is given by M_H I/SFR. The calculation of the dynamical mass and virial radius is described in Sect. 4.1.

The HALOGAS sample currently does not appear in a homogeneous sample of SFR determinations. The values listed here were predominantly compiled from a set of Hα and infrared surveys, allowing the determination of SFRs that are corrected for dust attenuation. The pilot paper did not list errors on the SFRs, and these errors can be significant because of the different methods, errors on the distances, or even the timescales probed in different bands. Here we followed Kennicutt et al. (2009), who determined that SFRs corrected for dust attenuation should differ by ±0.3 dex at most from their calculated value. If we assume this to correspond to a 3σ limit, the error on the values listed in Table 4 is 30% at most.

All-sky surveys such as WISE or DSS are not sufficiently sensitive to provide counterpart detections or stellar mass estimates for our SOF IA detections. This means that we cannot estimate the ratio of M_H I to M_* for the cloud candidates. We therefore relied on our NED search and our deep R-band images to verify that these cloud candidates are without an optical counterpart.

Finally, in order to ensure that our cloud candidates were not related to the Milky Way system of HVCs, we compared the location and systemic velocity of the targets to the Leiden/Argentine/Bonn Survey (LAB; Kalberla et al. 2005). We did not find any emission from the Galactic system that is near to any of our detections, neither in projected location on the sky nor in velocity. Therefore, we conclude that our HALOGAS findings are not caused by Galactic halo clouds that lie along the same line of sight.

4. Accretion rates

4.1. Accretion rate of the cloud candidates and companions

One of the goals of the HALOGAS survey is to determine the H I accretion rate in a sample of star-forming galaxies that spans a broad range of galaxy characteristics such as mass and environment. After we identified all H I sources around the sample galaxies in a systematic manner, we approximated the accretion rate for the full sample. In this section, we calculate the actual observed accretion rate for both the cloud candidates as well as all observed H I within the virial radius of the HALOGAS targets. To do this, we assumed that the cloud candidates fall freely onto the disk, and these rates are therefore upper limits.

To estimate the accretion rates due to H I clouds, we used the projected distance (d) that the cloud candidate needs to travel to be accreted onto the disk of the host galaxy and their freefall time (t_ff). The latter is calculated as

$$\begin{array}{c}\hfill t_{\mathit{ff}}=\frac{\pi }{2}\times \sqrt{\frac{d^3}{2G(M_{\mathrm{dyn}+m_{\mathrm{cloud}}})}},\end{array}$$(1)

with m_cloud the H I mass of the cloud candidate and M_dyn the dynamical mass, which was calculated using the line width W₂₀ and the spherical approximation as

$$\begin{array}{c}\hfill M_{\mathrm{dyn}}=\frac{{(0.5\times W_{20})}^2\times R}{\sin {(i)}^2\times \mathrm{G}},\end{array}$$(2)

with W₂₀ from our SOF IA runs and R half the major axis of the 3σ ellipse fit to the target by SOF IA. Even though these are not the most accurate estimates of the rotational velocity and size of the disk that are available, their accuracy is not expected to be a significant factor in the error on the estimated accretion rates. Using these values allows for a consistent approach for all galaxies in the sample. For this reason, we did not use the rotation curves from Marasco et al. (2019) either, as these are only derived for the intermediately inclined galaxies. The individual dynamical masses and inclinations for the targets are listed in Table 4.

The timescale for accretion, and thus the estimated accretion rate, depends on the trajectory of the assumed infall. We established a range by considering infall to the center (t_ff1) and to the edge of the galactic disk (t_ff2) as defined by the 3σ ellipse fit by SOF IA. The accretion rates are then $\dot{M}=\frac{m_{\mathrm{cloud}}}{t_{\mathit{ff}}}$ (see Table 5). The difference between the two calculated rates give an indication of the uncertainty of these accretion rates.

The average accretion rates of cloud candidates on each target galaxy are shown as circles in Fig. 2. In addition to calculating these accretion rates for our cloud candidates (Table 5), we also calculated the total accretion rates for all the targets, that is, including companions, in the sample. This was done in the same manner as for the cloud candidates, but now we only considered objects within the projected virial radius of the target, as defined by its dynamical mass, and for which the velocity difference between the source and host is smaller than the host escape velocity at the projected distance of the companion. The latter was derived from the host dynamical mass combined with the projected distance of the companions to the galaxy in the usual manner. These rates are shown in Fig. 2 as stars. The dashed line in this figure shows Ṁ = SFR, whereas the solid line shows Ṁ = 0.1 × SFR. These values simply set the range of timescales and are stringent upper limits on the accretion rates because we used projected distances and freefall times.

Fig. 2. Ṁ vs. SFR of host galaxies. Solid circles represent the average of the two calculated accretion rates, one with a timescale of tff1 (infall on center) and the other with a timescale of tff2 (infall on edge of disk), for galaxies with cloud candidates. The horizontal error indicates the distance to these upper limits. The errors on the SFR indicate the assumed 30% error. Stars show the same, but now for all sources around the target within the virial radius for the whole sample, barring the merging galaxies. The solid line represents Ṁ = 0.1 × SFR (galaxies to the right of this line are labeled with their name) and the dashed line Ṁ = SFR. The gray cross and plus symbol indicate where NGC 891 would fall according to our method and when individual SOF IA detections are not merged with the disk, respectively (see Sect. 5.2).

It is quite obvious from Fig. 2 that none of the galaxies here have an observed accretion rate anywhere near their SFRs when only the cloud candidates are considered. It even appears that the current observed neutral gas accretion rate is uncorrelated with the SFR. Even when we consider all observed gas within the virial radius, only one galaxy (NGC 4565) in the sample has sufficient H I in its surroundings to exceed its current SFR. Even considering the significant errors on the SFRs, no other galaxy would break this barrier. This implies that we do not fully trace the gas reservoir that fuels the star formation in these galaxies. Sancisi et al. (2008) found that the average total accretion rate amounted to ∼10% of the SFR, which broadly speaking, is consistent with the results presented in this paper. However, out of 20 galaxies in our sample, only 6 galaxies have observed accretion rate upper limits in excess of 10% of their SFR. They are labeled in Fig. 2. Simply assuming that the galaxies in the sample are all forming stars at some rate already poses a problem because in more than half our sample, we do not detect any H I within the virial radius, and in six observations, no H I is detected at all in addition to the target.

Calculating the accretion rate as described above, we find mean neutral gas accretion rates onto the disk of the host galaxies (Ṁ₂) of 0.03 M_⊙ yr⁻¹ for all cloud candidates, 0.18 M_⊙ yr⁻¹ for all companions within the virial radius, and 0.22 M_⊙ yr⁻¹ for all gas detected within the virial radius of our galaxies. Although the virial radius is likely to be an underestimate as we derived it from the dynamical mass in the galaxy, we find a negligible effect on the accretion rate estimates by increasing the virial radius by 50–100%, probably because in most cases, our sensitivity is severely reduced outside these radii.

These averages are consistent with previous results in the literature (e.g., Sancisi et al. 2008; Di Teodoro & Fraternali 2014). On the other hand, the average SFR in the HALOGAS sample (0.8 M_⊙ yr⁻¹) is slightly lower than that typically assumed or measured in spiral galaxies (Sancisi et al. 2008; Bothwell et al. 2011), but still well above the observed accretion rates.

We reiterate that these accretion rates are stringent upper limits because we used projected distances. For the total survey, we can obtain more realistic values for the accretion rates within the virial radius through a statistical correction. When we assume that the distance of the detection along the line of sight is half the distance to the boundary of the virial sphere on average, that is⁷ $d_{\mathrm{corr}}=\sqrt{\frac{3}{4}{d}^2+\frac{1}{4}{r}_{\mathrm{vir}}^2}$, we find an accretion rate of 0.05 (0.04) M_⊙ yr⁻¹ for Ṁ₂ (Ṁ₁), which increases the discrepancy with the average SFR.

Most of the sample-averaged accretion rate comes from a single galaxy, NGC 4565 (see Table 6), giving the impression that a significant part of accretion often occurs through minor mergers, while most of the time, little to no neutral gas is being accreted.

4.2. Upper limit on the accretion rate.

Although HALOGAS provides the deepest resolved survey of nearby galaxies to date, there is still potentially a large amount of undetected gas at low column density. Such a reservoir was recently claimed to have been detected in sensitive Green Bank Telescope (GBT) measurements for NGC 891 and NGC 4565 (Das et al. 2020). Here we attempt to estimate the neutral hydrogen gas reservoir that might still be hiding below our sensitivity limits. For this calculation, we deviated from the concept presented in the previous section, where we considered a proxy for the virial volume around our targets. Even though these volumes present a maximum amount of accretion, the timescales involved are long. As we wish to know whether the observed SFR can be fueled by H I accretion, in this section we consider an area closer to the disk to estimate the current accretion rate that can be hidden below our sensitivity limits.

In order to obtain a realistic estimate of the amount of gas that can fuel the current star formation, we need to estimate the time it takes a certain H I mass to cross the boundary of the star formation disk, that is, a crossing time (t_cross) and a total mass in the boundary area (M_boundary), such that

$$\begin{array}{c}\hfill {\dot{M}}_{\mathrm{upper}=\frac{M_{\mathrm{boundary}}}{t_{\mathrm{cross}}}},\end{array}$$(3)

with Ṁ_upper the upper limit accretion rate.

First we consider M_boundary. If we wish to estimate it from our detection limits, we require the number of clouds in the boundary region and multiply this number with their mass (M_cl). The mass is simply set by by the detection limit we obtained in Sect. 2. To estimate the number of clouds, we consider the volume of the boundary region to be two cylinders sitting right above and below the disk of the galaxies (Fig. 3), as defined by D₂₅ (see Table 4 and Heald et al. 2011b) and an arbitrary height h. Even though gas can be accreted outside the optical disk onto the H I disk, this gas would need to lose significant amounts of angular momentum before it would form stars. It would therefore only become available on timescales much longer than typically considered in star formation.

Fig. 3. Cartoon indicating two accreting clouds of radius a and infall velocity vcross in a volume of height h at an arbitrary distance from the midplane of the galaxy (here, D25 = 30 kpc).

The mass in this boundary volume (V_boundary) is then

$$\begin{array}{c}\hfill M_{\mathrm{boundary}}=\frac{V_{\mathrm{boundary}}}{V_{\mathrm{cl}}}\times {\mathrm{M}}_{\mathrm{cl}}\times f,\end{array}$$(4)

where f is a filling factor describing how much of the volume is filled with our clouds, and V_cl is the volume of our clouds. For V_cl, we consider spherical clouds with a diameter FWHM_maj. The filling factor for H I clouds is not well established. When we calculate this from the clouds we do detect within the total virial volume of the survey, we find f = 3 × 10⁻⁵. This is likely to be a gross underestimate as smaller clouds will be more numerous and thus take up a larger fraction of the considered volume. From the literature, a secure upper limit appears to be f = 0.1 − 0.2 (Dutta 2019, and references therein). Therefore, we used the range f = 3 × 10⁻⁵ − 0.1 as bracketing values for our upper limit on the accretion rates. For these calculations, we assumed that the product of mass of the cloud and the filling factor at a given cloud size is constant.

The time required to accrete this mass would be the time it takes the entire volume to sink into the disk, that is,

$$\begin{array}{c}\hfill t_{\mathrm{cross}}=\frac{h}{v_{\mathrm{cross}}},\end{array}$$(5)

with v_cross the speed of the accretion flow. Particles that are ejected from the disk in ballistic fountain models (Collins et al. 2002; Fraternali & Binney 2008) have typical vertical return speeds ∼50 − 100 km s⁻¹. In this calculation, we assume that the accretion of external gas occurs at similar speeds as the flows would mix, and the ballistic model provides an upper limit for clouds ejected from the disc. Because we are searching for an upper limit, we take the maximum speed, v_cross = 100 km s⁻¹, and we obtain an accretion rate of

$$\begin{array}{c}\hfill {\dot{M}}_{\mathrm{upper}}\le 3\frac{{D_{25}}^2}{a^3}\times {\mathrm{M}}_{\mathrm{cl}}\times f\times v_{\mathrm{cross}},\end{array}$$(6)

in which the height of the cylinders cancels out against the same height in the accretion time, and a is the diameter of the clouds under consideration, that is, FWHM_maj. As the accretion rate is a mass flux rate, the height of the cylinder should cancel out, and hence this height does not affect the final calculations. Using the previously stated range of f, this leads to an upper limit range on the accretion rate of 0.003–7.5 M_⊙ yr⁻¹ from the HALOGAS survey alone.

This range is not very informative because the filling factor is uncertain. Therefore, we turn to the aforementioned GBT observations. These observations hardly detect any additional H I compared to HALOGAS (Pingel et al. 2018; Pingel 2018), and gas that might be detected is at very low column densities (∼4 × 10¹⁷ cm⁻²; Das et al. 2020). Taking 5 × 10¹⁷ cm⁻² with a line width of 20 km s⁻¹ as an indicative upper limit on the detection from the GBT, we can obtain a limiting maximum mass by multiplying this limit with the area of the GBT beam and the mass of the hydrogen atom, which can still be present in the area of the GBT beam. We obtain the mass in our boundary area by replacing the cloud mass in Eq. (4) with the aforementioned limiting GBT mass and V_cl with the volume of the cylinder described by the GBT beam area with a depth of 2 × R_max (see Table 4). As the GBT volume is about V_boundary, we use f = 1 and obtain

$$\begin{array}{c}\hfill {\dot{M}}_{\mathrm{upper}}\le \frac{\pi }{4}\times {\sigma }_{N_{\mathrm{HI}}}\times m_{\mathrm{HI}}\times \sqrt{\frac{W_{20}}{20.}}\times \frac{D_{25}^2}{R_{\max }}\times v_{\mathrm{cross}}\end{array}$$(7)

as the upper limit on the accretion rate. In this, σ_NHI is the limiting column density, m_HI is the mass of a neutral hydrogen atom, and W₂₀ is the line width of the galaxy. The latter is to scale σ_NHI to the line width expected in the GBT cylinder under consideration (Wolfe et al. 2015). From this, we find an upper limit of 0.04 M_⊙ yr⁻¹, showing that the galaxies of the sample would not accrete enough neutral gas to maintain their SFR and are not embedded in extended neutral hydrogen gas reservoirs.

The values for the individual galaxies are shown in Fig. 4 and Table 7. The figure clearly shows that all accretion rate upper limits from the GBT limit are below the host SFR, and only for galaxies with a low SFR do the limits surpass 0.1 × SFR.

Fig. 4. Upper limits on Ṁ vs. SFR of host galaxies. Circles are as calculated with f = 0.1, stars with f = 3 × 10−5, and squares from scaling the GBT limits. Solid symbols represent hosts with cloud candidate detections, and open symbols show those without detections. The solid line indicates Ṁ = 0.1 × SFR, and the dashed line marks Ṁ = SFR.

4.3. Cloud kinematics

When investigating the individual fields, we marked every detection as corotating with or counter-rotating to the rotation in the disk of the target galaxy. A detection was considered corotating when its location and systemic velocity placed it on the same side of the central coordinates of the target as the disk rotation, and counter-rotating when this was not the case.

Figure 5 shows these kinematics for all detections in the sample. Here the cloud candidates (circles) and companions (stars) are separated, and we differentiate between sources within the virial radius (as defined in Sect. 4.1) of the target and outside this radius (small gray symbols). Inside the virial radius, companions and cloud candidates both appear to preferentially corotate. Sixteen out of 21 sources corotate. These 16 sources are almost equally split between cloud candidates and companions (9 and 7, respectively), making it likely that the cloud candidates and companions have a similar origin.

Fig. 5. Absolute velocity difference between the cloud or companion and the target, plotted against the projected distance of the cloud or companion from the target center. Open symbols indicate counter-rotating objects, and filled symbols indicate corotating objects. Large black symbols fall inside the host virial radius and small gray symbols outside the virial radius. Stars show companions, and circles show the cloud candidates.

5. Discussion

We have investigated many different parameters (number of group members and group luminosity (as listed in Heald et al. 2011b), SFR, depletion time, distance, mass, H I richness) of our sample to determine whether we might find any difference between the targets with cloud candidates and those without. However, none of these parameters showed significant evidence that the two groups come from different parent samples.

5.1. Comparison with HVCs

We now investigate whether our cloud candidates can be analogs to the HVCs found in the Milky Way and M 31. We compare their velocity width and peak column densities to observations of HVCs as observed by the H I Parkes All Sky Survey (HIPASS; Putman et al. 2002) in Fig. 6. Our detections occupy the same region in this diagram as the HVC population of the Milky Way. However, there are many more detections of clouds in the Milky Way than in our full sample. This is to be expected due to sensitivity effects and makes it questionable whether the peak column density is a useful comparison because it heavily depends on how well the cloud in question is resolved. A better comparison is made with the HVC mass.

Fig. 6. FWHM vs. peak column for the Milky Way HVCs observed in HIPASS (stars; Putman et al. 2002) and our detections (circles).

We compare our detections to the masses of the Milky Way HVC complexes, Complex C, Complex GCP, Complex WB, Chain A, and the Cohen Stream (van Woerden et al. 1999; Thom et al. 2006, 2008; Wakker et al. 2008), where we take the mass at the mean of the lower and upper distance limits. It is easier to obtain H I masses for clouds in external galaxies, therefore we added the cloud complex and HVC analogs that were found in the M 31 system (Thilker et al. 2004; Westmeier et al. 2008) to the mass comparison. This comparison is shown in Fig. 7. This figure shows that our detections are more massive than most of the detections in M 31 and the Milky Way. Additionally, far more clouds are detected around M 31 than in the galaxies in our sample. This is not simply a sensitivity issue, as we explain below.

Fig. 7. Mass vs. projected distance to the host for the clouds surrounding M 31 (stars; Westmeier et al. 2008) and our detections. Filled circles show the cloud candidates, and open circles show our detections with a stellar counterpart. Open squares indicate the Milky Way HVC complexes. Their errors indicate the actual observed ranges. The dashed, dot-dashed, and dotted lines indicate the minimum, median, and maximum detection limits in the survey galaxies. The limits are primary beam corrected and hence increase with increased distance to the host. The axes are not adjusted to show the companions, therefore only a few are visible.

The dashed, dot-dashed, and dotted lines in Fig. 7 indicate the minimum, median, and maximum detection limit in our sample of galaxies. The minimum and maximum correspond to galaxies NGC 4244 and UGC 7774, respectively, and the median limit was calculated using the median distance of 11.0 Mpc. The differences in sensitivity are predominantly due to the projected distance to the host and the corresponding primary beam correction when projected distances and fluxes are converted into physical units.

These sensitivity limits show that in our most sensitive observation, we expect to detect the majority of a population of clouds as seen in M 31. However, we find no cloud candidates in the cube for NGC 4244. In this specific case, this might still be due to imaging artifacts (see Appendix B.1) or low number statistics. However, in the two closest galaxies after NGC 4244 (NGC 5229 and NGC 5023), which therefore have the most sensitive detection limits in the center, no cloud candidates are detected either. We find only a few detections below the median detection limit.

Because we extensively tested our source-finding sensitivity limit in Sect. 2, it is unlikely that our sensitivity estimates should be increased by a factor of two. Figure 7 also shows that resolution is not an issue here because the projected distances of the M 31 sample and our detections are in a similar range. Furthermore, the four to five most massive clouds of the M 31 population fall securely in our detection range for several galaxies because detections for the sample overlap with clouds in M 31. When we consider the mass range 7 × 10⁵ M_⊙ < M_H I < 1.2 × 10⁶ M_⊙, we find two M 31 clouds and at least one Milky Way cloud. In the same mass bin in our sample, three cloud candidates are associated with three different galaxies. This means that we should have detected many more cloud candidates in several targets. However, we detect four candidates at most in a single galaxy.

In all targets of the survey, we are sensitive to clouds, such as the most massive H I complexes found in the Milky Way and M 31, that is, Complex C and the Davies Cloud (M_HI > 5×10⁶ M_⊙). However, we only detect five cloud candidates with a mass > 5 × 10⁶ M_⊙ in a sample of 20 galaxies, two of which are associated with interacting galaxies (see Appendix B.1.6). This lack of massive cloud complexes such as are present in the Milky Way and M 31 was previously reported toward the isolated galaxy NGC 2903 (Irwin et al. 2009), but is now confirmed for the first time in a sample of galaxies and shows that the distribution of HVCs around M 31 and the Milky Way is not typical for other spiral galaxies. This poses the question why our sample is so different from the Milky Way and M 31, or NGC 891 for that matter. In our sample, there is a minute indication that galaxies with more close-by companions have more cloud candidates, but the current sample is too small to draw any strong conclusions.

5.2. Implications on the cold-accretion mode

Observations have shown that the required ongoing gas supply of disk galaxies in the local Universe cannot be satisfied by minor mergers with gas-rich companions (this work; Sancisi et al. 2008; Di Teodoro & Fraternali 2014). Thus, accretion from intergalactic (primordial) gas potentially plays an important role in sustaining the SFRs of large disk galaxies. From a theoretical perspective, it is unclear whether this accretion should occur in a cold or hot mode in local galaxies (Nelson et al. 2013, 2015; Schaller et al. 2015; Huang et al. 2019). Therefore, collecting observational evidence for this accretion is crucial to understand galaxy evolution.

Fernández et al. (2012) investigated the H I distribution around a simulated Milky Way-sized galaxy and found that the neutral gas that is accreted consists of both compact clouds and filaments with peak column densities of ∼10¹⁹ cm⁻² out to at least 90 kpc from the accreting galaxy. Our results do not show any significant sign of ongoing accretion of cold neutral gas at these levels, although our data are sensitive to this. Recent GBT observations suggest that there is neutral gas at even lower column densities (Das et al. 2020). However, as shown in Sect. 4.2, this is not enough material to fuel the current SFRs. Hence, if the cold-mode accretion seen in the simulations exists in reality, it has to be in the ionized state.

The accretion rates, both observed and upper limits, presented in this paper rule out the possibility that galaxies are accreting neutral hydrogen gas at the same rate as they are forming stars. However, it is still necessary to confirm this in even more sensitive observations because our main conclusion comes with several caveats.

First, we did not identify accreting gas that is already connected to the main disk of its host. We have tested this by applying our source-finding method to NGC 891 (Oosterloo et al. 2007). Even though most of the gas previously identified is detected, the bulk of it is connected to the disk. This means that in our analysis, it would not be counted as accreting gas, but as part of the main disk. In the case of NGC 891, SOF IA initially detects the filament and some clouds as separate sources that are then grown to merge with the disk. When we do not combine these with the host, this would result in an additional 0.25 M_⊙ yr⁻¹ (see Fig. 2), a significant amount. However, a visual inspection of the data and masks in our sample shows only few such structures in the HALOGAS sample (see Appendix B.1 and Heald & HALOGAS Team, in prep.).

A second caveat is that our calculation assumes that the infall timescale is well characterized by the freefall time, which is likely to result in an underestimation of the timescale. Furthermore, the freefall time itself is likely to be overestimated because we calculated it from the dynamical mass as measured from the H I observation of the host galaxy (see Sect. 2). Hence, our observed accretion rate limits are uncertain in both directions.

The upper limits calculated in this paper were calculated by assuming that accreted gas only becomes available for star formation if it is accreted within the optical disk. This assumption was made because there are only few observations in which large-scale radial inflows are observed. Even in galaxies in which flows like this are detected (e.g., NGC 2403; Fraternali et al. 2002), the radial velocities of these flows are such that the migration of this gas to the inner disk would occur on timescales an order of magnitude longer than those sampled by the current star formation.

In our calculations, we assumed that the gas flows predominantly parallel to the angular momentum vector of the disks. This assumption ensures a high accretion rate, as any other path would be longer. The thick disks around intermediately inclined HALOGAS galaxies show radial inflows of about 20–30 km s⁻¹ (Marasco et al. 2019), which would result in longer infall times. On the other hand, the gas can be expected to follow the orientation of the magnetic fields in the halo (Kwak et al. 2009), which are often observed to have a significant component perpendicular to the disk (e.g., Stein et al. 2019).

For clarity, we did not incorporate a helium component in the considered masses. However, in the literature, gas masses are typically assumed to be 1.3–1.4 times the neutral hydrogen mass to account for the presence of helium. This correction would raise our combined average accretion limit to 0.13 M_⊙ yr⁻¹), still significantly below the average SFR in the sample (0.8 M_⊙ yr⁻¹). On average, the galaxies in this sample are observed to accrete only 9% of their current SFR in neutral gas when the accretion rates are corrected for the helium contribution. Another 7% of neutral hydrogen gas accretion might be hiding below the current GBT sensitivity limits. Finally, even though the HALOGAS collaboration has attempted to collect a set of SFRs as homogeneous and accurately as possible, it should be noted that SFRs for an individual galaxy are rather uncertain as they were derived from many different sources of information and include several assumptions and corrections.

The above discussion shows that the limits derived from the HALOGAS sample are, and can only be, rough estimates. Even though the survey shows that extended neutral gas reservoirs, such as found in NGC 891, M 31, or the Milky Way, are exceptional rather than commonplace, single-dish observations are still required to establish a meaningful upper limit on the possible remaining gas reservoir that can hide below the sensitivity limits. Future observations such as the MeerKAT HI Observations of Nearby Galactic Objects – Observing Southern Emitters (MHONGOOSE) Survey (de Blok et al. 2016) on MeerKAT will reach sensitivity limits on the column density equal to the current GBT observations with a spatial resolution that is about eight times higher. This will significantly improve the statistics and self-consistency of the observations and reduce uncertainties in the analysis presented here.

6. Conclusions

We have performed an automated search with SOF IA (Serra et al. 2015) of 22 galaxies in the HALOGAS survey (Heald et al. 2011b). A mock data cube has shown that with this method, we can detect sources down to a limit of 21 mJy km s⁻¹ integrated flux. Depending on the distance of the cloud, this translates into a detection threshold in the range ∼0.5–3 × 10⁶ M_⊙.

In our analysis, we found 14 detections for which we are unable to find an optical counterpart. These detections account for an average accretion rate of 0.03 M_⊙ yr⁻¹ over the whole sample. When we simply compare this to the average SFR in the sample, 0.8 M_⊙ yr⁻¹, this would not be sufficient to replenish the gas used in star formation. When we consider all detections within the virial radius of our host galaxies, we confirm previous estimates from the literature and find a stringent mean upper limit of the neutral gas (H I) accretion rate onto the disks of the galaxies of 0.22 M_⊙ yr⁻¹. This upper limit is based on projected distances and can statistically be corrected to a limit for actual distances for the average. This correction lowers the observed accretion rate to 0.05 M_⊙ yr⁻¹.

Our cloud candidates have similar peak column densities and FWHM in their velocity distributions as the HVCs in our Galaxy. However, they are far less numerous. For the HVCs, this could be merely a sensitivity issue as it is difficult to estimate the masses of this population. However, when we compare our results to the HVC analogs found around M 31, we also find fewer cloud candidates than expected, indicating that this type of population is missing or much less abundant in the HALOGAS galaxies. From our limited sample size, we cannot identify a definite cause for this difference, but there are indications that galaxies with closer companions have more clouds. If this trend were confirmed in a larger sample, it is likely that our cloud candidates have a tidal origin and are not caused by primordial accretion. For now, the question remains open why some galaxies display an HVC population and others do not.

The HALOGAS observations are some of the most sensitive observations at these resolutions, and the upper limits set by these observations on neutral gas accretion exclude that all accretion occurs in the form of neutral hydrogen down to column densities of a few times 10¹⁹ cm⁻². Because of the uncertainties in the filling factor of the gas, these limits still allow for a significant amount of neutral hydrogen to remain undetected. However, current GBT limits are more stringent, and based on these, we find an upper limit on the still-undetected gas accretion rate of 0.04 M_⊙ yr⁻¹ for the HALOGAS sample. Combined with the observed accretion rate of other gas detections, this leads to a total possible accretion rate of neutral hydrogen of Ṁ = 0.09 M_⊙ yr⁻¹ or Ṁ = 0.13 M_⊙ yr⁻¹ when the presence of helium is accounted for.

Thus the neutral hydrogen gas accretion onto z = 0 galaxies is lower than their current SFR, and this probably means that this type of accretion no longer occurs on large scales, or that it occurs in another phase than the neutral gas phase. More importantly, however, the HALOGAS survey shows that most nearby galaxies are not embedded in an extended neutral hydrogen reservoir, as is detected around some galaxies, for example, NGC 891. The more sensitive observations of the MHONGOOSE project (de Blok et al. 2016) will be able to provide better insights into the role played by neutral hydrogen in the accretion of gas onto galaxies and why some galaxies are embedded in extended gas reservoirs while most are not.

Acknowledgments

The work at RUB is partially supported by the BMBF project 05A17PC2 for D-MeerKAT. N.H.R. further acknowledges support from the BMBF through the project D-LOFAR IV (FKZ: 05A17PC1). This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program grant agreement no. 882793, project name MeerGas. This research has made use of the VizieR catalog access tool, CDS, Strasbourg, France (DOI : https://doi.org/10.26093/cds/vizier). The original description of the VizieR service was published in A&AS 143, 23. This publication makes use of data products from the Wide-field Infrared Survey Explorer, which is a joint project of the University of California, Los Angeles, and the Jet Propulsion Laboratory/California Institute of Technology, funded by the National Aeronautics and Space Administration. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration

References

Appendix A: Source catalog

Appendix B: Source visualization

B.1. Overview of the target fields

B.1.1. NGC 0925

NGC 925 is a spiral galaxy with an extended gas distribution and an XUV disk. In addition to the target, we found two more H I sources. One detection (source 2) was found relatively close to the target galaxy and might be part of the extended low surface brightness disk. The projected distance to the center of the target galaxy is ∼ 23 kpc, and its systemic velocity only differs by 30 km s⁻¹ from NGC 925. We were unable to find an optical counterpart, and hence it was classified as a cloud candidate. Technically, the cloud is counter-rotating. However, it is very close to the kinematic minor axis of the galaxy, making the kinematical assessment difficult.

The other detection (source 4) is separated from the target galaxy. It has a UV counterpart, as stated in Karachentsev et al. (2013), and our deep R-band image also shows an optical counterpart (see Figure B.30). Its systemic velocity deviates by about 90 km s⁻¹ from NGC 925, and it is at a projected distance of ∼ 62 kpc from the center of NGC 925 and corotates with its disk.

B.1.2. NGC 1003

This is an edge-on galaxy whose H I disk extends far beyond the visible optical disk. The outer parts of the disk show indications of warping, and several clouds, considered to be HVC analogs, can be seen around the main disk (Heald et al. 2011a). These clouds are all detected, but only one of them is separated from the emission in the disk (see section 5.2 for the effect on our results).

Even though this counter-rotating source is separated by more than one FWHM from the mask encompassing the emission of NGC 1003, it is still close to the disk. The HALOSTARS image (Figure B.17) does not show an optical counterpart at its location, but the field is crowded with foreground stars, making an identification of low-level optical emission difficult.

B.1.3. NGC 2541

A companion is found at a projected distance of 120 kpc with a velocity difference of only ∼40 km s⁻¹ (see Figure B.31). Even though the galaxy is warped, it appears to be rather regular and undisturbed. The detected companion is at the edge of the cube, and hence its detection is too faint in H I to provide any meaningful analysis of its structure or kinematics.

B.1.4. NGC 3198

This galaxy was extensively investigated for its halo properties in Gentile et al. (2013). It is a rather regular galaxy, in which we detect a single companion galaxy at a projected distance of ∼ 140 kpc. The systemic velocity of this companion is aligned with the rotation of the NGC 3198 disk.

B.1.5. NGC 4258

The field of NGC 4258 is a rich field with several detections. The galaxy has four smaller companions, which makes past or ongoing weak interaction very likely. In addition to the four companions, the galaxy has several tails that extend from the main disk. They typically end in fragmented clouds. Sometimes, they are connected to the disk and merged with the galaxy in our source finding (see Figure B.5 in the SE). In other instances, as in the case of our sources 3 and 8, they are distinct enough to be counted as individual detections. However, they all appear to be connected to the disk in some shape or form. It is therefore impossible to tell whether they are infalling or are tidally stripped from the galaxy. All these detections follow the general rotation of the disk. Sources 9 and 11 are far beyond the velocity range of the disk, however.

For source 10, we were unable to spectroscopically confirm an optical counterpart, but our KPNO R -band image B.36 shows a smudge of light. Therefore, it was classified as a companion. This source lies close to the minor axis of the host and has a systemic velocity ∼75 km s⁻¹, which is higher than the highest velocity in the disk of NGC 4258. Technically, this means that it is counter-rotating.

B.1.6. NGC 4274

NGC 4274 is another rich field with at least three companion galaxies, NGC 4286, NGC 4283, and NGC 4278. All are about 90 kpc south of the target and much closer to each other than to NGC 4274. Their systemic velocities and locations are aligned with the rotation in the disk of NGC 4274. The H I in NGC 4278 is sporadic and fragmented. However, this appears to be more an effect of sensitivity than the actual distribution. Some H I is detected toward NGC 4283, but it appears to be part of foreground emission coming from NGC 4278.

We find three cloud candidates in this field. Two (sources 9 and 10) are closer to the system of companion galaxies and are most likely tidal debris of interactions between these galaxies. Source 10 would be counter-rotating when compared to the disk of NGC 4274, but is corotating with the companion galaxies, and source 9 corotates with NGC 4274.

Source 11 is close to NGC 4274, is extended and has a hint of rotation. Unfortunately, the spatial resolution of the HALOGAS observations is not sufficient to confirm this rotation. Even in deep-follow-up observations there appears to be no optical counterpart, and hence we refer to this object as the dark galaxy candidate. If it truly is a galaxy, possibly with an undetected optical counterpart, it would be a companion and not a cloud. This cloud candidate is counter-rotating from the perspective of the NGC 4274 disk.

B.1.7. NGC 4414

The single cloud candidate detected in this cube appears at the end of a tail of the target galaxy and is corotating. In a galaxy with a low inclination like NGC 4414, this is likely material falling into the disk and not ejected from it, as most of the ejection mechanisms work in a vertical direction compared to the disk orientation. In addition to the detected cloud candidate, the galaxy also displays a significant extension in H I to the northwest. In our source finding, this has merged with the main disk. NGC 4414 is the galaxy with the highest SFR in the sample, and the HALOGAS data were previously described in de Blok et al. (2014).

In addition to the target and cloud candidate, we also detect NGC 4359. This galaxy is significantly removed from the target, and its systemic velocity is misaligned with the rotation of the NGC 4414 disk.

Fig. B.1. HALOSTARS image for the NGC 0925 field overlaid with contours from the detected sources. Contour colors correspond to the systemic velocity of each source as determined by SOF IA and indicated by the color bar on the right. Contour levels vary per source and are listed in the captions of the individual images. The contours for the target correspond to 5.1, 20.5, 40.9, 81.8, 122.7, and 162.3×1019cm−2. The solid blue ellipse indicates the nominal virial radius (which can fall outside the field of view), and the dashed ellipse shows the FWHM of the WSRT primary beam. Labels indicate the source number and class as listed in Tables 5 and A.1. Blue crosses indicate galaxies that are spectroscopically confirmed to be in the velocity range of the cube.

B.1.8. NGC 4448

NGC 4448 is an isolated, regularly rotating galaxy seen at a high inclination. The H I distribution in the galaxy appears to be rather compact although the line width is average (see Table 4). The optical disk extends beyond the H I disk, which gives the impression that the gas in the outer regions is being stripped. The galaxy IC 3334 is also detected in the data cube, but with a significantly higher systemic velocity.

B.1.9. NGC 4559

The HALOGAS data for NGC 4559 were presented and discussed in detail in Vargas et al. (2017). This is a moderately inclined galaxy without obvious peculiarities. We detect a single companion at a projected distance of 65 kpc. The companion shows no regular velocity structure, and the H I appears to be more extended than the optical from the WISE images. The difference in systemic velocity between this companion and NGC 4559 is more than 400 km s⁻¹.

B.1.10. NGC 4565

NGC 4565 is a disturbed galaxy with three companions. We find two anomalous gas clouds in this system that are close to the galaxy, and we classify them as cloud candidates. Additionally, we find another two cloud candidates in this data cube. All companions and cloud candidates are corotating with the disk of the galaxy, except for source 8, which is farthest removed in velocity as well as projected distance.

Fig. B.2. As Figure B.1, but for the NGC 1003 field.

Tidal debris is a possible origin for the corotating cloud candidates, given the environment of interaction. This observation was extensively described in Zschaechner et al. (2012).

B.1.11. NGC 4631

NGC 4631 is a galaxy that interacts heavily with its environment. This extended gas distribution was already seen in previous observations (Rand 1994), and, as there is gas almost everywhere in the field of view, is complicated to "clean". Hence, significant side lobes of the dirty beam remain in the final data cube, and we detect several artifacts. Complicating matters even more, it appears that several real sources are situated on top of side-lobe emission. They can be visually identified as they are so much brighter than the side lobes, but any velocity and flux measurements are highly unreliable.

In addition to the obvious companion NGC 4656, we detect three companions that spatially coincide with the gas surrounding NGC 4631, but are distinct in velocity space. Additionally, we detect a source that is further removed from the system. The latter (source 9) is the only companion that is not aligned with the rotational direction of the disk. As the galaxy is in an equal-mass merger and due to the issues with the side lobes, we excluded it from the analysis presented in this paper.

Fig. B.3. As Figure B.1, but for the NGC 2541 field.

B.1.12. NGC 5055

This is a large spiral galaxy. In addition to the target, we detect two companions, UGC 8313 and UGC 8365, and a gas cloud candidate that almost connects to the main disk. The cloud candidate and UGC 8365 follow the velocity orientation of the disk, although the velocity difference between NGC 5055 and UGC 8365 is so large that it is doubtful that they are in the same dynamical system. The systemic velocity of UGC 8313 is such that it coincides with the velocities on the receding side of NGC 5055, while its location is on the approaching side.

B.1.13. NGC 5585

This galaxy is a spiral galaxy with an H I disk that extends significantly beyond the optical detection in the DSS image and that is slightly warped. In addition to the target, we detect a single cloud candidate at a projected distance of 58 kpc. The cloud candidate shows an indication of a velocity structure, but as it is barely resolved, this might be an artifact of the masking. If this cloud were part of the dynamical system describing NGC 5585, it would be counter-rotating.

B.1.14. UGC 4278

In addition to the target galaxy, we find one other source in the field that we deem to be real, the companion NGC 2537. We do not detect anything in between the two sources, and their disks appear regular. We therefore assume that there is currently no interaction between the galaxies. If they were dynamically connected, their velocity structure would be such that NGC 2537 would be counter-rotating in the frame of UGC 4278, but in the frame of NGC 2357, the target would be corotating.

Fig. B.4. As Figure B.1, but for the NGC 3198 field.

B.1.15. UGC 7774

UGC 7774 is an edge-on galaxy with one of the most striking known examples of a warp (see Figure B.15). We detect a single companion at a rather large distance from the galaxy (d_proj∼ 232 kpc). This companion does appear quite disturbed, and hence it might be related to the warp in UGC 7774, but a detailed analysis would be required to further this idea. Its systemic velocity is aligned with the rotation on the western side of UGC 7774.

Fig. B.5. As Figure B.1, but for the NGC 4258 field overlaid on a KPNO R-band image.

Fig. B.6. As Figure B.1, but for the NGC 4274 field.

Fig. B.7. As Figure B.1, but for the NGC 4414 field.

Fig. B.8. As Figure B.1, but for the NGC 4448 field overlaid on a DSS 2 R-band image.

Fig. B.9. As Figure B.1, but for the NGC 4559 field overlaid on a KPNO R-band image.

Fig. B.10. As Figure B.1, but for the NGC 4565 field.

Fig. B.11. As Figure B.1, but for the NGC 4631 field overlaid on a KPNO R-band image.

Fig. B.12. As Figure B.1, but for the NGC 5055 field.

Fig. B.13. As Figure B.1, but for the NGC 5585 field overlaid on a KPNO R-band image.

Fig. B.14. As Figure B.1, but for the UGC 4278 field.

Fig. B.15. As Figure B.1, but for the UGC 7774 field overlaid on a DSS 2 R-band image.

B.2. Cloud candidates

Fig. B.16. Overview panels for source 2 in the cube of NGC 0925. From left to right: Optical image overlaid with contours of the H I moment 0 (intensity) map, moment 1 (velocity ) map, the PV diagram along the major axis, and line profile. Contours for moment 0 are 6.8, 27.2, and 54.4×1018cm−2 overlaid on our HALOSTARS R-band image. They are corrected with the same factor as the integrated flux in the various tables to account for the primary beam response. The blue ellipse indicates the SOF IA fit ellipse. Contours for the velocity field start at 560.36 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 206° with contours at -3, -1.5, 1.5, and 3 σ, with σ = 0.160 mJy beam−1.

Fig. B.17. As B.16, but for source 2 in the cube of NGC 1003. Contours for moment 0 are 7.7, 31.0, and 62.0×1018cm−2. Contours for the velocity field start at 571.56 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 123° with contours at -1.5, 1.5, and 3 σ, with σ = 0.182 mJy beam−1.

Fig. B.18. As B.16 but for source 3 in the cube of NGC 4258. Contours for moment 0 are 1.2, and 5.0×1019cm−2 but overlaid on our KPNO R-band image. Contours for the velocity field start at 238.00 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 148° with contours at -1.5, 1.5, and 3 σ, with σ = 0.215 mJy beam−1.

Fig. B.19. As B.16 but for source 8 in the cube of NGC 4258. Contours for moment 0 are 1.6 and 6.5×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 555.24 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 349° with contours at -1.5, 1.5, and 3 σ, with σ = 0.227 mJy beam−1.

Fig. B.20. As B.16 but for source 9 in the cube of NGC 4274. Contours for moment 0 are 1.1, 4.5, and 9.0×1019cm−2. Contours for the velocity field start at 781.04 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 186° with contours at -1.5 and 1.5 σ, with σ = 0.173 mJy beam−1.

Fig. B.21. As B.16 but for source 10 in the cube of NGC 4274. Contours for moment 0 are 8.3, 33.1, and 66.3×1018cm−2. Contours for the velocity field start at 962.32 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 206° with contours at -1.5, 1.5, and 3σ, with σ = 0.161 mJy beam−1.

Fig. B.22. As B.16 but for source 11 in the cube of NGC 4274. Contours for moment 0 are 5.3, 21.2, 42.5, and 63.7×1018cm−2. Contours for the velocity field start at 978.80 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 71° with contours at -1.5, 1.5, 3, and 6 σ, with σ = 0.172 mJy beam−1.

Fig. B.23. As B.16 but for source 2 in the cube of NGC 4414. Contours for moment 0 are 6.1 and 24.2×1018cm−2. Contours for the velocity field start at 609.16 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 139° with contours at -1.5, 1.5, and 3 σ, with σ = 0.189 mJy beam−1.

Fig. B.24. As B.16 but for source 3 in the cube of NGC 4565. Contours for moment 0 are 5.8, 23.1, and 46.2×1018cm−2. Contours for the velocity field start at 996.40 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 13° with contours at -1.5, 1.5, and 3 σ, with σ = 0.170 mJy beam−1.

Fig. B.25. As B.16 but for source 6 in the cube of NGC 4565. Contours for moment 0 are 5.0, 19.9, and 39.8×1018cm−2. Contours for the velocity field start at 1317.76 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 341° with contours at -1.5, 1.5, 3, and 6 σ, with σ = 0.156 mJy beam−1.

Fig. B.26. As B.16 but for source 7 in the cube of NGC 4565. Contours for moment 0 are 7.3 and 29.0×1018cm−2. Contours for the velocity field start at 1363.08 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 5° with contours at -1.5, 1.5, and 3 σ, with σ = 0.211 mJy beam−1.

Fig. B.27. As B.16 but for source 8 in the cube of NGC 4565. Contours for moment 0 are 9.6, 38.5, 77.0, and 153.9×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 1527.88 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 186° with contours at -1.5, 1.5, and 3 σ, with σ = 0.193 mJy beam−1.

Fig. B.28. As B.16 but for source 2 in the cube of NGC 5055. Contours for moment 0 are 1.2, 4.8, 9.6, and 14.4×1019cm−2. Contours for the velocity field start at 511.92 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 209° with contours at -3, -1.5, 1.5, and 3 σ, with σ = 0.187 mJy beam−1.

Fig. B.29. As B.16, but for source 1 in the cube of NGC 5585. Contours for moment 0 are 6.1, 24.2, and 48.4×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 78.12 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 135° with contours at, 1.5, 3, and 6 σ, with σ = 0.274 mJy beam−1.

B.3. Companions

Fig. B.30. Moment maps for source 4 in the cube of NGC 0925. Contours for moment 0 are 1.9, 7.6, 15.2, 30.4, and 45.6×1019cm−2 overlaid on our HALOSTARS R-band image. They are corrected with the same factor as the integrated flux in the various tables to account for the primary beam response. The blue ellipse indicates the SoFiA fit ellipse. Contours for the velocity field start at 622.16 km s−1 and increase with 5.00 km s−1.

Fig. B.31. As B.30, but for source 1 in the cube of NGC 2541. Contours for moment 0 are 1.4, 5.5, 11.0, and 16.5×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 494.44 km s−1 and increase with 5.00 km s−1. This source is identified as WISEA J081239.49+483645.3.

Fig. B.32. As B.30 but for source 2 in the cube of NGC 3198. Contours for moment 0 are 1.1, 4.2, 8.5, 16.9, 25.4, and 33.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 543.76 km s−1 and increase with 5.00 km s−1. This source is identified as VV 834 NED02.

Fig. B.33. As B.30 but for source 4 in the cube of NGC 4258. Contours for moment 0 are 3.4 and 13.7×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 349.24 km s−1 and increase with 5.00 km s−1.

Fig. B.34. As B.30 but for source 6 in the cube of NGC 4258. Contours for moment 0 are 1.3, 5.3, and 10.5×1020cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 423.40 km s−1 and increase with 5.77 km s−1. This source is identified as NGC 4248.

Fig. B.35. As B.30 but for source 9 in the cube of NGC 4258. Contours for moment 0 are 2.2, 8.8, 17.6, 35.2, and 52.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 823.04 km s−1 and increase with 19.98 km s−1.

Fig. B.36. As B.30 but for source 10 in the cube of NGC 4258. Contours for moment 0 are 1.2, 4.8, and 9.6×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 798.32 km s−1 and increase with 5.00 km s−1.

Fig. B.37. As B.30 but for source 11 in the cube of NGC 4258. Contours for moment 0 are 1.1, 4.3, and 8.6×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 930.16 km s−1 and increase with 5.00 km s−1.

Fig. B.38. As B.30 but for source 1 in the cube of NGC 4274. Contours for moment 0 are 1.6, 6.6, 13.1, and 19.7×1019cm−2. Contours for the velocity field start at 377.28 km s−1 and increase with 23.90 km s−1.

Fig. B.39. As B.30 but for source 6 in the cube of NGC 4274. Contours for moment 0 are 1.5, 6.2, 12.4, 24.7, and 37.1×1019cm−2. Contours for the velocity field start at 575.04 km s−1 and increase with 6.59 km s−1. This source is identified as NGC 4286.

Fig. B.40. As B.30 but for source 7 in the cube of NGC 4414. Contours for moment 0 are 1.2, 4.8, 9.6, 19.2, and 28.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 1136.52 km s−1 and increase with 10.92 km s−1. This source is identified as NGC 4359.

Fig. B.41. As B.30 but for source 2 in the cube of NGC 4448. Contours for moment 0 are 4.0, 15.9, 31.8, and 63.6×1019cm−2, but overlaid on a DDS 2 R-band image. Contours for the velocity field start at 1186.52 km s−1 and increase with 5.00 km s−1. This source is identified as IC 3334.

Fig. B.42. As B.30 but for source 2 in the cube of NGC 4559. Contours for moment 0 are 2.4, 9.4, 18.9, 37.7, and 56.6×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 1172.48 km s−1 and increase with 5.00 km s−1. This source is identified as WISEA J123521.10+273342.9.

Fig. B.43. As B.30 but for source 1 in the cube of NGC 4565. Contours for moment 0 are 5.9, 23.6, 47.2, and 94.3×1018cm−2. Contours for the velocity field start at 839.84 km s−1 and increase with 5.00 km s−1.

Fig. B.44. As B.30 but for source 4 in the cube of NGC 4565. Contours for moment 0 are 6.7, 26.8, 53.6, 107.1, and 160.7×1019cm−2. Contours for the velocity field start at 1260.08 km s−1 and increase with 8.45 km s−1. This source is identified as NGC 4562.

Fig. B.45. As B.30 but for source 5 in the cube of NGC 4565. Contours for moment 0 are 5.4, 21.6, and 43.3×1019cm−2. Contours for the velocity field start at 1223.00 km s−1 and increase with 5.00 km s−1. This source is identified as IC 3571.

Fig. B.46. As B.30 but for source 2 in the cube of NGC 4631. Contours for moment 0 are 7.1, 28.6, and 57.2×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 514.72 km s−1 and increase with 12.57 km s−1. This source is identified as NGC 4656.

Fig. B.47. As B.30 but for source 6 in the cube of NGC 4631. Contours for moment 0 are 1.4 and 5.4×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 646.56 km s−1 and increase with 5.00 km s−1. This source is identified as SDSS J124146.99+325124.8.

Fig. B.48. As B.30 but for source 9 in the cube of NGC 4631. Contours for moment 0 are 2.9 and 11.5×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 712.48 km s−1 and increase with 5.56 km s−1. This source is identified as SDSS J124010.08+323930.4.

Fig. B.49. As B.30 but for source 10 in the cube of NGC 4631. Contours for moment 0 are 10.0, 40.0, and 79.9×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 852.56 km s−1 and increase with 5.77 km s−1. This source is identified as MCG +06-28-022.

Fig. B.50. As B.30 but for source 11 in the cube of NGC 4631. Contours for moment 0 are 7.3, 29.1, 58.2, and 116.3×1018cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 889.64 km s−1 and increase with 5.00 km s−1.

Fig. B.51. As B.30 but for source 4 in the cube of NGC 5055. Contours for moment 0 are 3.2, 12.9, and 25.9×1020cm−2. Contours for the velocity field start at 549.00 km s−1 and increase with 7.42 km s−1. This source is identified as UGC 8313.

Fig. B.52. As B.30 but for source 6 in the cube of NGC 5055. Contours for moment 0 are 1.3, 5.3, and 10.7×1020cm−2. Contours for the velocity field start at 1187.60 km s−1 and increase with 7.21 km s−1. This source is identified as UGC 8365.

Fig. B.53. As B.30 but for source 4 in the cube of UGC 4278. Contours for moment 0 are 1.3, 5.1, and 10.2×1020cm−2. Contours for the velocity field start at 349.88 km s−1 and increase with 9.27 km s−1. This source is identified as NGC 2537.

Fig. B.54. As B.30, but for source 2 in the cube of UGC 7774. Contours for moment 0 are 2.2, 8.9, and 17.8×1020cm−2, but overlaid on a DDS 2 R-band image. Contours for the velocity field start at 632.24 km s−1 and increase with 5.00 km s−1. This source is identified as MCG +07-26-024.

All Tables

All Figures

	Fig. 1. Integrated flux (circles, mJy km s−1, right and bottom axes) and spectral peak flux densities (stars, mJy, left and top axes) from the input catalog vs. SOF IA-retrieved values for the detected sources. Black and gray are the sources detected by SOF IA, red is the false detection.
In the text

Fig. 2. Ṁ vs. SFR of host galaxies. Solid circles represent the average of the two calculated accretion rates, one with a timescale of tff1 (infall on center) and the other with a timescale of tff2 (infall on edge of disk), for galaxies with cloud candidates. The horizontal error indicates the distance to these upper limits. The errors on the SFR indicate the assumed 30% error. Stars show the same, but now for all sources around the target within the virial radius for the whole sample, barring the merging galaxies. The solid line represents Ṁ = 0.1 × SFR (galaxies to the right of this line are labeled with their name) and the dashed line Ṁ = SFR. The gray cross and plus symbol indicate where NGC 891 would fall according to our method and when individual SOF IA detections are not merged with the disk, respectively (see Sect. 5.2).

In the text

	Fig. 3. Cartoon indicating two accreting clouds of radius a and infall velocity vcross in a volume of height h at an arbitrary distance from the midplane of the galaxy (here, D25 = 30 kpc).
In the text

	Fig. 4. Upper limits on Ṁ vs. SFR of host galaxies. Circles are as calculated with f = 0.1, stars with f = 3 × 10−5, and squares from scaling the GBT limits. Solid symbols represent hosts with cloud candidate detections, and open symbols show those without detections. The solid line indicates Ṁ = 0.1 × SFR, and the dashed line marks Ṁ = SFR.
In the text

Fig. 5. Absolute velocity difference between the cloud or companion and the target, plotted against the projected distance of the cloud or companion from the target center. Open symbols indicate counter-rotating objects, and filled symbols indicate corotating objects. Large black symbols fall inside the host virial radius and small gray symbols outside the virial radius. Stars show companions, and circles show the cloud candidates.

In the text

	Fig. 6. FWHM vs. peak column for the Milky Way HVCs observed in HIPASS (stars; Putman et al. 2002) and our detections (circles).
In the text

Fig. 7. Mass vs. projected distance to the host for the clouds surrounding M 31 (stars; Westmeier et al. 2008) and our detections. Filled circles show the cloud candidates, and open circles show our detections with a stellar counterpart. Open squares indicate the Milky Way HVC complexes. Their errors indicate the actual observed ranges. The dashed, dot-dashed, and dotted lines indicate the minimum, median, and maximum detection limits in the survey galaxies. The limits are primary beam corrected and hence increase with increased distance to the host. The axes are not adjusted to show the companions, therefore only a few are visible.

In the text

Fig. B.1. HALOSTARS image for the NGC 0925 field overlaid with contours from the detected sources. Contour colors correspond to the systemic velocity of each source as determined by SOF IA and indicated by the color bar on the right. Contour levels vary per source and are listed in the captions of the individual images. The contours for the target correspond to 5.1, 20.5, 40.9, 81.8, 122.7, and 162.3×1019cm−2. The solid blue ellipse indicates the nominal virial radius (which can fall outside the field of view), and the dashed ellipse shows the FWHM of the WSRT primary beam. Labels indicate the source number and class as listed in Tables 5 and A.1. Blue crosses indicate galaxies that are spectroscopically confirmed to be in the velocity range of the cube.

In the text

	Fig. B.2. As Figure B.1, but for the NGC 1003 field.
In the text

	Fig. B.3. As Figure B.1, but for the NGC 2541 field.
In the text

	Fig. B.4. As Figure B.1, but for the NGC 3198 field.
In the text

	Fig. B.5. As Figure B.1, but for the NGC 4258 field overlaid on a KPNO R-band image.
In the text

	Fig. B.6. As Figure B.1, but for the NGC 4274 field.
In the text

	Fig. B.7. As Figure B.1, but for the NGC 4414 field.
In the text

	Fig. B.8. As Figure B.1, but for the NGC 4448 field overlaid on a DSS 2 R-band image.
In the text

	Fig. B.9. As Figure B.1, but for the NGC 4559 field overlaid on a KPNO R-band image.
In the text

	Fig. B.10. As Figure B.1, but for the NGC 4565 field.
In the text

	Fig. B.11. As Figure B.1, but for the NGC 4631 field overlaid on a KPNO R-band image.
In the text

	Fig. B.12. As Figure B.1, but for the NGC 5055 field.
In the text

	Fig. B.13. As Figure B.1, but for the NGC 5585 field overlaid on a KPNO R-band image.
In the text

	Fig. B.14. As Figure B.1, but for the UGC 4278 field.
In the text

	Fig. B.15. As Figure B.1, but for the UGC 7774 field overlaid on a DSS 2 R-band image.
In the text

Fig. B.16. Overview panels for source 2 in the cube of NGC 0925. From left to right: Optical image overlaid with contours of the H I moment 0 (intensity) map, moment 1 (velocity ) map, the PV diagram along the major axis, and line profile. Contours for moment 0 are 6.8, 27.2, and 54.4×1018cm−2 overlaid on our HALOSTARS R-band image. They are corrected with the same factor as the integrated flux in the various tables to account for the primary beam response. The blue ellipse indicates the SOF IA fit ellipse. Contours for the velocity field start at 560.36 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 206° with contours at -3, -1.5, 1.5, and 3 σ, with σ = 0.160 mJy beam−1.

In the text

	Fig. B.17. As B.16, but for source 2 in the cube of NGC 1003. Contours for moment 0 are 7.7, 31.0, and 62.0×1018cm−2. Contours for the velocity field start at 571.56 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 123° with contours at -1.5, 1.5, and 3 σ, with σ = 0.182 mJy beam−1.
In the text

	Fig. B.18. As B.16 but for source 3 in the cube of NGC 4258. Contours for moment 0 are 1.2, and 5.0×1019cm−2 but overlaid on our KPNO R-band image. Contours for the velocity field start at 238.00 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 148° with contours at -1.5, 1.5, and 3 σ, with σ = 0.215 mJy beam−1.
In the text

	Fig. B.19. As B.16 but for source 8 in the cube of NGC 4258. Contours for moment 0 are 1.6 and 6.5×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 555.24 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 349° with contours at -1.5, 1.5, and 3 σ, with σ = 0.227 mJy beam−1.
In the text

	Fig. B.20. As B.16 but for source 9 in the cube of NGC 4274. Contours for moment 0 are 1.1, 4.5, and 9.0×1019cm−2. Contours for the velocity field start at 781.04 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 186° with contours at -1.5 and 1.5 σ, with σ = 0.173 mJy beam−1.
In the text

	Fig. B.21. As B.16 but for source 10 in the cube of NGC 4274. Contours for moment 0 are 8.3, 33.1, and 66.3×1018cm−2. Contours for the velocity field start at 962.32 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 206° with contours at -1.5, 1.5, and 3σ, with σ = 0.161 mJy beam−1.
In the text

	Fig. B.22. As B.16 but for source 11 in the cube of NGC 4274. Contours for moment 0 are 5.3, 21.2, 42.5, and 63.7×1018cm−2. Contours for the velocity field start at 978.80 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 71° with contours at -1.5, 1.5, 3, and 6 σ, with σ = 0.172 mJy beam−1.
In the text

	Fig. B.23. As B.16 but for source 2 in the cube of NGC 4414. Contours for moment 0 are 6.1 and 24.2×1018cm−2. Contours for the velocity field start at 609.16 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 139° with contours at -1.5, 1.5, and 3 σ, with σ = 0.189 mJy beam−1.
In the text

	Fig. B.24. As B.16 but for source 3 in the cube of NGC 4565. Contours for moment 0 are 5.8, 23.1, and 46.2×1018cm−2. Contours for the velocity field start at 996.40 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 13° with contours at -1.5, 1.5, and 3 σ, with σ = 0.170 mJy beam−1.
In the text

	Fig. B.25. As B.16 but for source 6 in the cube of NGC 4565. Contours for moment 0 are 5.0, 19.9, and 39.8×1018cm−2. Contours for the velocity field start at 1317.76 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 341° with contours at -1.5, 1.5, 3, and 6 σ, with σ = 0.156 mJy beam−1.
In the text

	Fig. B.26. As B.16 but for source 7 in the cube of NGC 4565. Contours for moment 0 are 7.3 and 29.0×1018cm−2. Contours for the velocity field start at 1363.08 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 5° with contours at -1.5, 1.5, and 3 σ, with σ = 0.211 mJy beam−1.
In the text

	Fig. B.27. As B.16 but for source 8 in the cube of NGC 4565. Contours for moment 0 are 9.6, 38.5, 77.0, and 153.9×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 1527.88 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 186° with contours at -1.5, 1.5, and 3 σ, with σ = 0.193 mJy beam−1.
In the text

	Fig. B.28. As B.16 but for source 2 in the cube of NGC 5055. Contours for moment 0 are 1.2, 4.8, 9.6, and 14.4×1019cm−2. Contours for the velocity field start at 511.92 km s−1 and increase with 5.00 km s−1. The PV Diagram is extracted at a PA of 209° with contours at -3, -1.5, 1.5, and 3 σ, with σ = 0.187 mJy beam−1.
In the text

	Fig. B.29. As B.16, but for source 1 in the cube of NGC 5585. Contours for moment 0 are 6.1, 24.2, and 48.4×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 78.12 km s−1 and increase with 5.00 km s−1. The PV diagram is extracted at a PA of 135° with contours at, 1.5, 3, and 6 σ, with σ = 0.274 mJy beam−1.
In the text

Fig. B.30. Moment maps for source 4 in the cube of NGC 0925. Contours for moment 0 are 1.9, 7.6, 15.2, 30.4, and 45.6×1019cm−2 overlaid on our HALOSTARS R-band image. They are corrected with the same factor as the integrated flux in the various tables to account for the primary beam response. The blue ellipse indicates the SoFiA fit ellipse. Contours for the velocity field start at 622.16 km s−1 and increase with 5.00 km s−1.

In the text

	Fig. B.31. As B.30, but for source 1 in the cube of NGC 2541. Contours for moment 0 are 1.4, 5.5, 11.0, and 16.5×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 494.44 km s−1 and increase with 5.00 km s−1. This source is identified as WISEA J081239.49+483645.3.
In the text

	Fig. B.32. As B.30 but for source 2 in the cube of NGC 3198. Contours for moment 0 are 1.1, 4.2, 8.5, 16.9, 25.4, and 33.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 543.76 km s−1 and increase with 5.00 km s−1. This source is identified as VV 834 NED02.
In the text

	Fig. B.33. As B.30 but for source 4 in the cube of NGC 4258. Contours for moment 0 are 3.4 and 13.7×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 349.24 km s−1 and increase with 5.00 km s−1.
In the text

	Fig. B.34. As B.30 but for source 6 in the cube of NGC 4258. Contours for moment 0 are 1.3, 5.3, and 10.5×1020cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 423.40 km s−1 and increase with 5.77 km s−1. This source is identified as NGC 4248.
In the text

	Fig. B.35. As B.30 but for source 9 in the cube of NGC 4258. Contours for moment 0 are 2.2, 8.8, 17.6, 35.2, and 52.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 823.04 km s−1 and increase with 19.98 km s−1.
In the text

	Fig. B.36. As B.30 but for source 10 in the cube of NGC 4258. Contours for moment 0 are 1.2, 4.8, and 9.6×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 798.32 km s−1 and increase with 5.00 km s−1.
In the text

	Fig. B.37. As B.30 but for source 11 in the cube of NGC 4258. Contours for moment 0 are 1.1, 4.3, and 8.6×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 930.16 km s−1 and increase with 5.00 km s−1.
In the text

	Fig. B.38. As B.30 but for source 1 in the cube of NGC 4274. Contours for moment 0 are 1.6, 6.6, 13.1, and 19.7×1019cm−2. Contours for the velocity field start at 377.28 km s−1 and increase with 23.90 km s−1.
In the text

	Fig. B.39. As B.30 but for source 6 in the cube of NGC 4274. Contours for moment 0 are 1.5, 6.2, 12.4, 24.7, and 37.1×1019cm−2. Contours for the velocity field start at 575.04 km s−1 and increase with 6.59 km s−1. This source is identified as NGC 4286.
In the text

	Fig. B.40. As B.30 but for source 7 in the cube of NGC 4414. Contours for moment 0 are 1.2, 4.8, 9.6, 19.2, and 28.8×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the HALOSTARS field. Contours for the velocity field start at 1136.52 km s−1 and increase with 10.92 km s−1. This source is identified as NGC 4359.
In the text

	Fig. B.41. As B.30 but for source 2 in the cube of NGC 4448. Contours for moment 0 are 4.0, 15.9, 31.8, and 63.6×1019cm−2, but overlaid on a DDS 2 R-band image. Contours for the velocity field start at 1186.52 km s−1 and increase with 5.00 km s−1. This source is identified as IC 3334.
In the text

	Fig. B.42. As B.30 but for source 2 in the cube of NGC 4559. Contours for moment 0 are 2.4, 9.4, 18.9, 37.7, and 56.6×1019cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 1172.48 km s−1 and increase with 5.00 km s−1. This source is identified as WISEA J123521.10+273342.9.
In the text

	Fig. B.43. As B.30 but for source 1 in the cube of NGC 4565. Contours for moment 0 are 5.9, 23.6, 47.2, and 94.3×1018cm−2. Contours for the velocity field start at 839.84 km s−1 and increase with 5.00 km s−1.
In the text

	Fig. B.44. As B.30 but for source 4 in the cube of NGC 4565. Contours for moment 0 are 6.7, 26.8, 53.6, 107.1, and 160.7×1019cm−2. Contours for the velocity field start at 1260.08 km s−1 and increase with 8.45 km s−1. This source is identified as NGC 4562.
In the text

	Fig. B.45. As B.30 but for source 5 in the cube of NGC 4565. Contours for moment 0 are 5.4, 21.6, and 43.3×1019cm−2. Contours for the velocity field start at 1223.00 km s−1 and increase with 5.00 km s−1. This source is identified as IC 3571.
In the text

	Fig. B.46. As B.30 but for source 2 in the cube of NGC 4631. Contours for moment 0 are 7.1, 28.6, and 57.2×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 514.72 km s−1 and increase with 12.57 km s−1. This source is identified as NGC 4656.
In the text

	Fig. B.47. As B.30 but for source 6 in the cube of NGC 4631. Contours for moment 0 are 1.4 and 5.4×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 646.56 km s−1 and increase with 5.00 km s−1. This source is identified as SDSS J124146.99+325124.8.
In the text

	Fig. B.48. As B.30 but for source 9 in the cube of NGC 4631. Contours for moment 0 are 2.9 and 11.5×1020cm−2, but overlaid on a DDS 2 R-band image as it is outside the KPNO field. Contours for the velocity field start at 712.48 km s−1 and increase with 5.56 km s−1. This source is identified as SDSS J124010.08+323930.4.
In the text

	Fig. B.49. As B.30 but for source 10 in the cube of NGC 4631. Contours for moment 0 are 10.0, 40.0, and 79.9×1019cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 852.56 km s−1 and increase with 5.77 km s−1. This source is identified as MCG +06-28-022.
In the text

	Fig. B.50. As B.30 but for source 11 in the cube of NGC 4631. Contours for moment 0 are 7.3, 29.1, 58.2, and 116.3×1018cm−2, but overlaid on our KPNO R-band image. Contours for the velocity field start at 889.64 km s−1 and increase with 5.00 km s−1.
In the text

	Fig. B.51. As B.30 but for source 4 in the cube of NGC 5055. Contours for moment 0 are 3.2, 12.9, and 25.9×1020cm−2. Contours for the velocity field start at 549.00 km s−1 and increase with 7.42 km s−1. This source is identified as UGC 8313.
In the text

	Fig. B.52. As B.30 but for source 6 in the cube of NGC 5055. Contours for moment 0 are 1.3, 5.3, and 10.7×1020cm−2. Contours for the velocity field start at 1187.60 km s−1 and increase with 7.21 km s−1. This source is identified as UGC 8365.
In the text

	Fig. B.53. As B.30 but for source 4 in the cube of UGC 4278. Contours for moment 0 are 1.3, 5.1, and 10.2×1020cm−2. Contours for the velocity field start at 349.88 km s−1 and increase with 9.27 km s−1. This source is identified as NGC 2537.
In the text

	Fig. B.54. As B.30, but for source 2 in the cube of UGC 7774. Contours for moment 0 are 2.2, 8.9, and 17.8×1020cm−2, but overlaid on a DDS 2 R-band image. Contours for the velocity field start at 632.24 km s−1 and increase with 5.00 km s−1. This source is identified as MCG +07-26-024.
In the text