FEASTS Combined with Interferometry. I. Overall Properties of Diffuse H i and Implications for Gas Accretion in Nearby Galaxies

We present a statistical study of the properties of diffuse H i in 10 nearby galaxies, comparing the H i detected by the single-dish telescope FAST (FEASTS program) and the interferometer Very Large Array (THINGS program), respectively. The THINGS observation missed H i with a median of 23% due to the short-spacing problem of interferometry and limited sensitivity. We extract the diffuse H i by subtracting the dense H i, which is obtained from the THINGS data with a uniform flux-density threshold, from the total H i detected by FAST. Among the sample, the median diffuse-H i fraction is 34%, and more diffuse H i is found in galaxies exhibiting more prominent tidal-interaction signatures. The diffuse H i we detected seems to be distributed in disk-like layers within a typical thickness of 1 kpc, different from the more halo-like diffuse H i detected around NGC 4631 in a previous study. Most of the diffuse H i is cospatial with the dense H i and has a typical column density of 1017.7–1020.1 cm−2. The diffuse and dense H i exhibit a similar rotational motion, but the former lags by a median of 25% in at least the inner disks, and its velocity dispersions are typically twice as high. Based on a simplified estimation of circumgalactic medium properties and assuming pressure equilibrium, the volume density of diffuse H i appears to be constant within each individual galaxy, implying its role as a cooling interface. Comparing with existing models, these results are consistent with a possible link between tidal interactions, the formation of diffuse H i, and gas accretion.


Introduction
In modern astrophysics, galaxies and their environments are considered an ecosystem of dark matter, stars and their remnants, and gaseous components (Naab & Ostriker 2017).Gas flows link the different parts of this system, cause matter/ momentum/energy to exchange, and thus drive the growth and evolution of galaxies.The details of how these processes work remains a major question in galaxy astrophysics (Somerville & Davé 2015).From large to small scales, the key processes include gas-removing environmental effects, gas accretion into circumgalactic medium (CGM) from the intergalactic medium (IGM), gas condensation into star-forming disks from the CGM, the flow of gas within the disk, the fueling of gas to star formation and black hole growth, and the feedback afterwards (Crain & van de Voort 2023).A complete observational view of different scales provides key constraints on galaxy formation models.
A long-lasting obstacle in modeling the processes of gas flowing through galaxies is the multiphase nature of gas.Localized instabilities easily develop in a multiphase gas and can then propagate and grow, or dissipate and fade in ways that are not well understood and are highly stochastic (Gronke et al. 2022).Multiwavelength observations from X-ray to radio are necessary to trace the multiphase, dynamic picture of baryonic flow all the way from the IGM to the star-forming regions.The neutral atomic hydrogen gas (H I) is an important part of the multiphase gas at a wide range of scales.In the interstellar medium (ISM) where it is abundant, H I is the raw material to form molecular hydrogen that eventually leads to star formation (Saintonge & Catinella 2022).But even in CGM and IGM, where the H I is insignificant in mass budget (Tumlinson et al. 2017), its amount, distribution, and kinematics shed light on the complex gas physics there.Examples include the H I tidal structures possibly inducing gas cooling through turbulent mixing (Sparre et al. 2022;Wang et al. 2023), the predicted small H I cloudlets in the 10 7 K CGM of elliptical galaxies formed through thermal instabilities (Nelson et al. 2020), the observed high-velocity H I clouds around the Milky Way (MW), marking the tip of the huge iceberg of the warm ionized gas being accreted onto the disk (Richter 2017), and the observed thick H I disk representing the interface of CGM condensing into the ISM (Marasco et al. 2019).Detecting and characterizing the H I in different gaseous environments provide useful clues to galaxy evolution.It is worth pointing out that, so far, most quantifications of H I with column densities less than or equal to 10 19 cm −2 are based on Lyα absorption or Lyman limit systems.
Traditional interferometric radio telescopes have been very helpful in imaging the H I in nearby galaxies through the 21 cm emission lines with a resolution sufficient to resolve starforming clumps, but they struggle in simultaneously capturing the more diffuse and extended H I. This is because interferometers construct images of the sky by filtering structures of specific spatial scales, and are thus limited by uv coverage of the shortest baselines.For the time being, the most promising way to solve the problem is using large single-dish radio telescopes to fill in the "short spacings," in order to recover the large-scale missing H I (Stanimirovic 2002).Recent technical advances include Kurono et al. (2009), Koda et al. (2011), and Rau et al. (2019), and recent scientific applications in this direction include Hess et al. (2017), de Blok et al. (2018), Richter et al. (2018), Das et al. (2020), Eibensteiner et al. (2023), and Wang et al. (2023).Most of these scientific studies confirm that interferometers tend to miss H I flux, though the exact amounts depend on source structures and telescope details.
We have been conducting an observing program FAST Extended Atlas of Selected Targets Survey (FEASTS; PIs: Jing Wang & Jie Wang) at the Five-hundred-meter Aperture Spherical radio Telescope (FAST) since 2021, to map the H I in the disk as well as the ∼100 kpc surroundings of nearby galaxies.With FEASTS, the H I gas is mapped with 21 cm emission lines down to a few times 10 17 cm −2 column densities, approaching regimes typically detected by Lyman limit systems.The first FEASTS science was conducted on the H I in/around the interacting galaxy NGC 4631 (Wang et al. 2023, hereafter W23).That work identified a component of excess (by 40%) H I detected by FAST but missed by Westerbork Synthesis Radio Telescope (WSRT) as part of the Hydrogen Accretion in LOcal GAlaxieS (HALOGAS) survey (Heald et al. 2011), which, until recently, was the deepest interferometric H I survey.We refer to this excess H I as the diffuse H I, because it has a large scale (∼30 kpc), moderate column density (10 20 cm −2 ), and high-velocity dispersion (∼50 km s −1 ).The diffuse H I is more closely related to the warm dense H I that has a velocity dispersion larger than 8 km s −1 than to the kinematically cooler dense H I, and has a column density high enough to induce cooling flows of the hot CGM.Put together, the diffuse H I seems to serve as an intermediate phase between the dense H I and warm ionized gas, and thus traces any energy inputting or dissipating processes that transfer gas between the cool (neutral) and hot (ionized) phases.In the context of galaxy evolution, the energy input or dissipative processes include tidal interaction, spiral arm and bar dynamics, stellar feedback, and black hole feedback.The dissipating or cooling processes includes gas accretion into the ISM, condensation of atomic to molecular phase, and fueling of gas to star formation and black hole growth.These are major physical processes in a typical galaxy formation model driving the baryonic flow and galaxy mass growth and evolution.The diffuse H I newly detected thus provides a promising tool to understand both gas physics and galaxy evolution.
Among the many physical problems of galaxy evolution, how galaxies replenish their gas to sustain star formation has been one of the most challenging.Theoretically, galactic gas accretion channels include cosmological hot and cold mode accretion, stripping from or merger of satellite galaxies, and fountain circulation (Putman 2017).Cosmological cold mode accretion through filaments penetrating the CGM is predicted to be significantly weakened at low redshift around massive galaxies with masses close to or above the MW (Dekel & Birnboim 2006).Cosmological hot mode accretion tends to couple with the latter two channels, as the circulated gas and stripped gas mix with, enrich, and accelerate cooling of the pristine CGM gas in the accretion process (Grand et al. 2019).The significance of each channel has been only loosely constrained from observations, but it has become observationally feasible to search for and characterize signatures of gas accretion through the fountain and tidal/merger channels (Fraternali 2017).HALOGAS has greatly advanced the frontier of quantifying extraplanar H I closely related with the fountain mechanism (e.g., Marasco et al. 2019).The diffuse H I detected far from the galactic disk of NGC 4631 suggests that the CGM cooling induced by tidal tails could be very effective (W23).
In this work, the first of a series of studies combining FEASTS with interferometry, we extend the work of W23, and look into a sample of 10 nearby galaxies with both single-dish H I images from FEASTS and interferometric H I images from The H I Nearby Galaxy Survey (THINGS; Walter et al. 2008, hereafter W08).The major science goal of this paper is to combine the THINGS and FEASTS data to get a first-order measure of the diverse properties of diffuse H I in general H I-rich galaxies, including its amount, morphology, distribution, and kinematics.As elaborated in the literature (e.g., Stanimirovic 2002), a combined analysis fully exploiting the advantages of both datasets is complex.In the first place, we need to ensure the flux calibration between the two datasets is consistent, so a technical goal of this paper is to optimize a simple procedure of flux cross-calibration in the image domain.After reasonable cross-calibration of fluxes, we simply define the diffuse H I as the excess H I detected by FAST in comparison to uniformly thresholded interferometric data to get a quick overview of the properties of this gas component.The definition will improve based on astrophysical arguments in the future, by filtering linearly combined or jointly deconvolved images and cubes (X.Lin et al. 2024, in preparation; J. Wang et al. 2024, in preparation).We also use relatively simple ways to quantify localized properties of the diffuse H I, but keep in mind that the structure and kinematics of the diffuse H I can be complex, which requires sophisticated decompositions in the future (X.Lin et al. 2024, in preparation;S. Oh et al. 2024, in preparation).We will show that even the coarse measurements already point to the interesting and different nature of diffuse H I compared to dense H I; they represent an ensemble measure of complex neutral gas structures, and theoretically may well link to large-scale CGM condensation processes (Gaspari et al. 2018).
The paper is organized as follows.We introduce the sample and data in Section 2. We describe the observational setting and data reduction procedure of the FEASTS data and a re-CLEAN of the THINGS data.Section 3 describes the data analysis.We cross-calibrate the World Coordinate System (WCS) systems and flux intensity levels for the two datasets, which are important steps before scientifically combining the two types of data.The strategies of the flux cross-calibrating procedure are optimized through mock tests presented in Appendix C. We then make the data cubes and moment images for the diffuse H I. The results are presented in Section 4. We show how much H I is missed in interferometric observations, and major dependence of the integral diffuse H I fraction.The column density distributions of different types of H I are displayed.The moment images of the diffuse H I are inspected, and radial profiles of properties measured from them are investigated and compared with those of dense H I. The possible physical meanings of the diffuse H I measurements are discussed in Section 5. We speculate the possible origins of the diffuse H I, including fountains and tidal interactions, the possible link of diffuse H I to gas accretion, and the difference and similarities between the diffuse H I detected here and that around NGC 4631 in W23.Conclusions are made in Section 6.All image data used and radial distribution measurements discussed will be published online.14

The Sample
The sample targeted in this study is a subset from THINGS (W08).We start from the subsample of 17 spiral galaxies used by Bigiel et al. (2010) to study the star-forming efficiency of H I gas in galactic outer disks.We exclude four galaxies at sky locations beyond an efficient observation of FAST, which have a decl.above 60°or below −5°.We further exclude three galaxies that have relatively small H I disks (R 4 H I < ¢) compared to the FAST beam FWHM of 3 24.The R HI is the semimajor axis of the isophote where the H I surface density reaches 1 M e pc −2 .
The remaining 10 galaxies are the target sample of this study (Table 1).These galaxies all have high-quality and relatively uniform interferometric 21 cm H I data taken at the Very Large Array (VLA) by the THINGS team, which is convenient for combining with the FAST H I image data.Scientific analysis of the combined H I data in this and future studies will benefit from the rich multiwavelength data, including deep mid-and far-infrared images from the Spitzer Infrared Nearby Galaxies Survey (Kennicutt et al. 2003) and Herschel Key Insights on Nearby Galaxies: A Far-Infrared Survey with Herschel (Kennicutt et al. 2011), deep ultraviolet images from Galaxy Evolution Explorer (GALEX) Nearby Galaxy Survey (NGS; Gil de Paz et al. 2007), images of total CO (2-1) from HERA CO-Line Extragalactic Survey (Leroy et al. 2009), images of dense CO (2-1) from Physics at High Angular resolution in Nearby GalaxieS (PHANGS) Atacama Large Millimeter/ submillimeter Array (Leroy et al. 2021), integral field spectroscopy data from PHANGS-Multi Unit Spectroscopic Explorer (Emsellem et al. 2022), etc.We utilize measurements of the central velocity, distance, inclination, and optical radius R 25 of the galaxies from W08. R 25 is the semimajor axis of the 25 mag arcsec −2 isophote in the B band.We measure the characteristic radius R H I of the H I disks from the THINGS column density images, using the procedure of Wang et al. (2016).We take the Wide-field Infrared Survey Explorer (WISE) W1, W2, and W4 fluxes, and GALEX far-ultraviolet (FUV) fluxes from the z = 0 Multiwavelength Galaxy Synthesis (Leroy et al. 2019).We estimate the stellar mass (M * ) with the W1 and W2 luminosities, using the equation 7 of Querejeta et al. (2015).We estimate the star formation rate (SFR) with the FUV and W4 luminosities, using the equation 1.11 of Calzetti (2013).We list these properties in Table 1.

FEASTS Data
The FAST H I image observations were mostly conducted in the years 2022 and 2023, the first and second observing years of the FEASTS.The observing project IDs are PT2021_0071 and PT2022_0096, respectively.The observations have been conducted in a mode similar to basket weaving, using  horizontal and vertical on-the-fly scans.The receiver is rotated by 23°. 4 and 53°. 4 in horizontal and vertical scans, respectively, to achieve an effective sampling spacing of 1.15', so as to satisfy the Nyquist-Shannon criterion.We refer readers to W23 for more details of the observational settings.We provide observing information for each target in Table 2, including the observing date, the median zenith angle, the southeast and northwest corners of the scanning region, the effective integration time per pointing, and the radio frequency interference (RFI) contaminating fraction.
The median zenith angles are all below 30°, causing only a low level of gain attenuation, which we use the equation 5 from Jiang et al. (2020) to correct for in data reduction.The RFI contaminating rate is minimal.The observation of NGC 5457 was the most strongly affected by RFI, but after inspecting the data, we confirm that the RFI did not significantly contaminate the region of the galaxy.
The data are reduced by a pipeline developed by the FEASTS team (W23).The pipeline includes standard steps of RFI flagging, calibration, gridding, and continuum subtraction, and is optimized for the FEASTS data.We extract from the raw data a frequency slice of 10 MHz (equivalent to ∼2000 km s −1 ) to focus on the surrounding environment of each target galaxy.As in W23, for each observed rectangular region, we slightly remove data on the four sides in the gridding procedure, to ensure a relatively uniform sampling density throughout the image.The cutout has a typical size of 1 deg 2 .Other details of the procedures and parameter settings can be found in W23.The product for each target is a data cube with a spatial resolution of 3 24, pixel size of 30″, and channel width of 1.61 km s −1 .A flux detection mask is generated for each data cube through SoFiA (Serra et al. 2015).When running SoFiA on FEASTS images, we use the smooth and clip finder with a smoothing kernel of 0, 3, 5, and 7 pixels along the x and y directions, and 0, 3, 7, and 13 channels along the z direction.The large extent of smoothing along the z-axis is motivated by the detection of diffuse H I with high levels of velocity dispersion around the galaxy NGC 4631 in W23.We set the detection threshold to be 4σ and the reliability threshold to be 0.99.
We estimate the rms level of each data cube from the blank regions beyond the SoFiA masks.We list the rms level and the related 3σ column density limit of H I by assuming a line width of 20 km s −1 in Table 2.We make moment 0, 1, and 2 maps based on the SoFiA mask of each target source.We show the column density, velocity, and velocity dispersion images based on these maps in the first row of Figures 13-22 in Appendix F. We also present an atlas of false-color images in Figure 1, to give a quick impression of how the FEASTS-detected H I is distributed with respect to the optical part of the galaxies and surroundings.

THINGS Data
THINGS H I data were taken at the VLA in its B, C, and D array configurations, achieving baselines that range from 35 to 11.4 km, corresponding to a spatial resolution of 25 2-4 64.The publicly available cubes of W08 were reduced by the THINGS team with the Astronomical Image Processing System.Conventionally, standard cubes are produced by directly adding residual cubes to the clean-beam convolved model cubes (convolved model cubes for short hereafter).But in W08, the residual maps are rescaled to account for the difference of areas between the dirty and clean beams.This algorithm has been discussed in detail in Jorsater & van Moorsel (1995) and Walter & Brinks (1999).The rescaled residual cubes are added to the convolved model cubes to produce rescaled cubes.
We do not use the cubes and related products from W08, including moment images, spectra, and integral fluxes, for all the targets in our sample.We redo the CLEAN procedure, starting from the calibrated and continuum removed visibilities obtained in W08, for three major reasons.First, the outer regions of a few galaxies with extended H I structures (NGC 628, NGC 5055, NGC 5194, NGC 5457) have been excluded in the W08 cube creation to avoid dealing with regions with significant primary beam attenuation, and to achieve a relatively uniform noise level throughout the image.For a complete comparison (and feathering in the future) with the FEASTS data, we require better coverage of the outer regions.Second, compared to the classical CLEAN algorithm used in W08, the multiscale CLEAN may work better at recovering the large-scale fluxes.Third, it may be useful to test the possibility that the convolved model cubes work better than the rescaled cubes in various analyses combined with singledish data.
We use the CASA script tclean to conduct the imaging and CLEAN procedures for all the galaxies except for NGC 7331.We use a pixel size of 1 5 following the data release of W08, "natural" weighting to maximize sensitivity, a cleaning threshold of 1.5σ, and a maximum cleaning iteration of 500,000.We do not use the robust weighting that usually produces better-shaped dirty beams because the sensitivity will be worse while the analysis combined with the FEASTS data is strongly limited by the interferometric depth (Sections 4.3, 4.4.2, and Appendix C).We use the "multiscale" deconvolver with a small scale bias of 0.2, and scales of roughly 0, 1, 2, 4, and 8 times the beam major FWHM.Through inspecting convolved models and residual images, we find that the multiscale CLEAN may not be the best deconvolver for NGC 7331, possibly because extended fluxes are hidden by the relatively high noise level and high inclination angle of the galaxy.Instead, the hybrid Hogbom/ Clark/Steer CLEAN algorithm of Miriad script clean produces a reasonably good cube for this galaxy.The CLEAN threshold and iterations are set as for the other galaxies.
We derive the source masks from standard cubes using SoFiA.We use the threshold finder, with smoothing kernels of sizes of 30″ along the x and y directions, and three channels along the z direction.We use a detection threshold of 2σ and a reliability threshold of 0.99.These settings largely follow those of W08.For each target, we use the region within the SoFiA mask to derive the scaling factor to correct for difference between dirty and clean beams, following the method of W08.We rescale the residual cubes before adding them to the convolved model cubes, and produce the rescaled cubes.We then use the SoFiA masks to derive moment maps from the rescaled cubes.The related column density, velocity, and velocity dispersion images are displayed in the second row of Figures 13-22 in Appendix F.
We list in Table 3 the rms level, the beam size, and H I column density limit of each standard cube.There might be systematic differences between the new cubes and the W08 cubes due to different reduction processes.We will compare measurements from the two sets of cubes to assess the systematic difference in most analyses.
We have used the W08 method to scale the CLEAN residuals before adding them to the convolved models.The alternative method to deal with residual fluxes of the CLEAN process is to deeply clean the data with a "clean mask."This method first identifies regions where deconvolution is need, and then within these regions cleans down very deeply to a fraction of the rms (e.g., 0.1σ).This avoids the scaling of residuals.We have adopted the W08 method instead of the deep clean, in consideration of the low signal-to-noise ratio (S/N) and large angular-scale nature of the THINGS data.The detailed justification can be found in Appendix A.

Analysis
In the following, we cross-calibrate the geometries and fluxes between the THINGS and FEASTS data, and analyze the possible excess H I detected by FEASTS in comparison to the THINGS data.We refer to the directly measured excess of H I from FEASTS compared to the THINGS data as the missed H I. We set a uniform threshold to select dense H I from the THINGS data, and quantify the respective excess H I from FEASTS data as the diffuse H I.

Aligning the FEASTS and THINGS Images
The FAST observations have a typical pointing uncertainty of 10″-20″, so we align the FAST images to the THINGS ones before conducting the flux cross-calibration.The pointing accuracy should not influence the calibration by comparing amplitudes of Fourier transformed images, but affect the sanity check using the imaginary and real parts (Section 3.2).It also significantly influences the deviation of moment images for the diffuse H I (Section 3.3).
We conduct the alignments by deriving the shifts needed in the x and y directions for the FEASTS and THINGS moment 0 images to be closest near the galaxy center.In practice, we modify the header of the FEASTS moment 0 image by adding values ranging from −0.5 to 0.5 pixels (−15 to 15″) with a step of 0.1 pixel (3″) to the CRPIX1 and CRPIX2 keywords separately.We reproject each shifted FEASTS moment 0 image to the THINGS moment 0 image through comparing their WCS keywords.We convolve the THINGS moment 0 image with the FAST beam and subtract it from the reprojected FEASTS moment 0 image.We calculate the standard deviation of fluxes within a box with widths of 300″ around the galaxy center from the difference image.The set of x and y shifting values that result in the lowest standard deviation are taken as the best correction, and this is applied to the header of the FEASTS data.

Cross-calibrating the Fluxes between THINGS and FEASTS
An important step before we discuss excess or diffuse H I is to achieve accurate flux calibration of the FEASTS data using the THINGS data as the reference.In principle, the systematic difference of fluxes between different telescope observations should be at a relatively small level of a few percent.We conduct the flux cross-calibration through comparing amplitudes of fast Fourier transform (FFT) images in overlapping uv space of the two types of data.We present details of setting up    the cross-calibration procedure in Appendix B, which is improved based on the version of W23, and is adjusted to suit the THINGS data.We present the derived scaling factor f F/T in Table 4, by which the original FEASTS fluxes should divide in order to be consistent with the THINGS ones.The median f F/T is 1.07 ± 0.07, consistent with the fact that systematic uncertainties of the flux calibration in H I observations are typically a few percent (W08).We also derive the scaling factors with the moment 0 images of W08 for all 10 galaxies (also in Table 4).Because only the inner regions (covered by both types of cubes) are used for cross-calibration, the scaling factors from using the two sets of VLA cubes are expected to be close.We can see that the two types of f F/T are indeed largely consistent, with a median difference of 0.02 ± 0.06.The consistency in f F/T supports the robustness of our method of deriving the scaling factor between the FAST and VLA fluxes.All FEASTS H I fluxes are divided by the corresponding f F/T values from the newly reduced THINGS cubes, and all THINGS data are corrected for the primary beam attenuation hereafter.
In Appendix B, the comparison in amplitudes of FFT images also reveals that the FAST observations tend to reveal excess fluxes when the angular scale is larger than 8 82, corresponding to 18.7-37.7 kpc in the sample.It is thus reasonable to quantify the spatial distribution and kinematics of the missed H I and the closely related diffuse H I (Section 3.3) with the FAST resolution of 3 2.

Extracting the Missed H I and Diffuse H I
We obtain the cube of the missed H I for each galaxy as the FEASTS cube minus the resolution-degraded THINGS cube.The resolution-degraded THINGS cube is produced by reprojecting the original THINGS cube to the WCS coordinates of the corresponding FEASTS cube and then convolving the reprojected cube with the FEASTS beam.When the spatially resolved properties of the THINGS-detected H I are compared to those of the missed H I or total H I, they are always measured from the resolution-degraded THINGS data in this study.
The amount of missed H I sensitively depends on the varying detection limits of the THINGS data.So we extract the diffuse H I by first separating the dense H I with a relatively uniform flux density threshold from the THINGS data and then subtracting it from the total H I. For each THINGS cube, we set a uniform flux density threshold as b b 0.005 arcsec mJy beam , where b maj and b min are the major and minor axes of the THINGS synthesis beam (Table 1).We select voxels with flux densities above this threshold in each THINGS cube and produce a region mask.We expand the mask by two pixels and blank all THINGS cube voxels beyond the mask.By doing so, we have produced a data cube for the relatively dense H I, dense H I cubes for short hereafter.The threshold used corresponds to a 3σ H I column density limit of 1.83 × 10 20 cm −2 assuming a line width of 20 km s −1 , which is higher than the column density limits of original THINGS data for most galaxies except for NGC 925 and NGC 7331 (Table 3).But because source finding for the original THINGS cubes is based on smoothed data, the relatively low threshold used in these two galaxies does not significantly bias the detection of dense H I, which is later supported by the lower values of dense H I fluxes than the THINGS total H I fluxes for these two galaxies (Section 4.2).The dense H I cubes are reprojected to the WCS coordinates of the corresponding FEASTS cubes, and then convolved with the FEASTS beam.The column density, velocity, and velocity dispersion images from the degraded cubes of dense H I are displayed in the third row in Figures 13-22 in Appendix F. Hereafter, all spatial properties of the dense H I are measured from the resolution-degraded data.
The cube of diffuse H I for each galaxy is obtained as the FEASTS cube minus the degraded dense H I cube.The diffuse H I defined in this way is roughly the missed H I plus the H I that lies between the THINGS detection limits and the selection criteria described above from the THINGS data.It should be a mixture of low-density H I and large-scale H I. We derive moment 0, 1, and 2 images of the diffuse H I with the SoFiA masks of the FEASTS cubes.Only line of sights where the column densities of diffuse H I are above 10 18 cm −2 and the ratios of diffuse H I over dense H I higher than 1% are considered in the deviation of the moment 1 and 2 images.We display the related column density, velocity, and velocity dispersion images of the diffuse H I in the fourth row in Figures 13-22 in Appendix F. Obtaining the moment images of diffuse H I through projecting the diffuse H I cubes is totally equivalent to deriving them through mathematical combinations of moment images of the dense H I and total H I. We put the relations in Appendix E, which helps cross-check patterns from images of different H I components and assess errors propagated from the THINGS data.
We also derive the difference velocity image as the degraded dense H I moment 1 image minus the diffuse H I moment 1 image.The difference velocity images show how much slower or faster the rotational/orbital motion of the diffuse H I is compared to that of the dense H I. The difference velocity image for each galaxy is displayed in the fifth row of Figures 13-22 in Appendix F.
Except in Sections 4.1 and 4.3, we present results mostly on the diffuse H I in the main part of the paper.

Deriving Radial Profiles of the Diffuse H I Emission
We derive the radial distribution of properties for the total H I, the dense H I (degraded to FAST resolution), and the diffuse H I, which are denoted by the subscripts tot, dens, and diff, respectively.
For the moment 0 and 2 images of different H I types, we derive radial profiles using annuli that have the same position angles and axis ratios as the optical disks (Table 1).All profile measurements start from the FAST beam FWHM to minimize the spatial smoothing effects of the beam.We do not correct for projection effects, as the geometry of the diffuse H I in outer disks is highly uncertain.
We also derive radial profiles of the normalized velocity difference between the dense and diffuse H I. The normalized velocity difference measures the lagged extents of rotational/ orbital velocities of the diffuse H I in comparison to the dense H I. For each galaxy, a slit is put along the major axis of the disk, with a width equal to the FAST beam FWHM.The averaged values are obtained along this slit from the difference velocity image (v r,dens − v r,diff ) and from the velocity image of the dense H I (v r,dens ).The former is divided by the latter to obtain the normalized velocity difference.The normalized velocity difference is positive and within unity when the diffuse H I is corotating (or orbiting in the same direction) with but more slowly than the dense H I, is negative when the former is corotating (or orbiting in the same direction) with but faster than the latter, and above unity when the former is counterrotating (or orbiting in the opposite direction) with respect to the latter.The infalling gas from a companion coming in the same direction as the disk or from a well-aligned CGM should still orbit in the same direction, but if the interaction geometry is inclined or the CGM is misaligned, the infall gas forms a warped outer disk or tail, which has a chance to show (projected) faster or opposite orbiting.

Results
In the following, we present results of this paper.We briefly discuss the amount of missed H I, which is the H I detected by the FEASTS but not by the THINGS data.We focus more on the diffuse H I, which includes the missed H I but is more uniformly selected by flux densities than the missed H I.

Global Measures of the Missed H I
The integrated H I fluxes of FEASTS and the fraction of missed H I ( f missed ) in the THINGS data estimated from both newly reduced cubes and W08 cubes are listed in Table 5 (and also shown in Figure 2).The total H I fluxes and amount of flux underestimation in this THINGS subset is for the first time uniformly quantified with a homogeneous and deep single-dish image dataset.
Among the whole sample, f missed has a median value of 23%.A significant fraction of missed H I is detected in almost all galaxies except for NGC 2841 and NGC 3198, which have high inclinations.The negative f missed of NGC 3198 should reflect noise, and the low f missed of NGC 2841 should also indicate nondetection.The f missed ranges from 17% to 44% in the eight galaxies with detectable missed H I. These detections are robust against the systematic uncertainty of 5% related to the flux cross-calibration (Appendix C.1). Comparisons of H I spectra from the FEASTS and THINGS data and for the missed H I are displayed in Figure 3.They confirm the f missed values discussed above.
In Table 5 (and Figure 2), we also see that the f missed values based on newly reduced VLA cubes are comparable with those based on the W08 cubes.The two types of f missed on the median differ by 0.0% ± 4.3%.This reflects that the multiscale CLEAN used for most galaxies in this paper do not recover significantly more H I flux than those of W08, possibly because the data are more limited by high noise levels than by limited baseline short spacing.This result is different from a few studies focusing on other subsets of the THINGS sample (Rich et al. 2008), which almost always obtain more fluxes with multiscale CLEAN than with classical CLEAN.In addition to possibly different structures of galaxies and rms levels of data, we rescaled the residual cubes before adding them to the convolved model cubes while many of the other studies do not, which may also be relevant in explaining the apparent discrepancy.
In Figure 2, we show the relation between H I masses and stellar masses, with the H I masses derived from both THINGS and FEASTS data.For half of the sample, the difference in H I masses from the two types of data are comparable to the scatter  The integrated H I spectra from FEASTS and THINGS data.The red solid lines and blue dashed curves are for the FEASTS and THINGS spectra, respectively.The cyan dotted curves plot the spectra of the missed H I, which is the FEASTS spectrum minus the THINGS spectrum for each galaxy.The magenta dotted curves plot the spectra of the diffuse H I, which is the FEASTS spectrum minus the dense H I spectrum for each galaxy.
deviation of which is commonly interpreted as H I amount fluctuations due to star formation fueling and feedback (Ostriker & Kim 2022) or the lagged response of star formation to abrupt gas removal or accretion (Cortese et al. 2021).Thus, recovering the H I missed in the interferometric observation is essential to assess the evolutionary stage and link global and local properties of these nearby galaxies.
We expect that the face-on, large galaxies, which have large angular scales along the minor axes in H I, tend to have higher f missed , as confirmed by the trend of red triangles in Figure 4. Several observational effects play a role.When the H I distribution minor axis is small, the angular scales of H I column density variations are shifted to low values even for a large disk, so the interferometric short-spacing problem becomes less severe.Meanwhile, the column densities are higher in more inclined systems, so the sensitivity problem becomes less severe.These two effects are confirmed by the positive trend of f missed,PBa with the galaxy minor axis (green triangles in Figure 4), where f missed,PBa is similar to f missed but derived after we apply the VLA primary beam response to both FEASTS and THINGS H I images to control for the primary beam attenuation effect.Furthermore, when both major and minor axes are large, the H I in the outer region is hard to detect in the interferometry data due to the strong primary beam attenuation and rms-levelbased source finding.This effect is reflected by the offsets between the missed H I fraction with and without primary beam attenuation applied (f missed -f missed, PBa , vertical distance between green and red triangles for each galaxy in Figure 4).The offsets tend to be larger for the four largest galaxies.Primary beam attenuation only accounts for a moderate fraction (<1/6) of f missed for the two largest galaxies, and never exceeds half f missed for the whole sample.Due to these effects, it is more likely to have high values of f missed in face-on large galaxies.
Finally, we note that five galaxies from the sample are also included in the Arecibo Legacy Fast ALFA (ALFALFA) Catalog for Extended Sources (Hoffman et al. 2019).The ALFALFA flux over FEASTS flux ratios range from 0.90 to 1.02, with a median value of 0.92.The on average slightly lower ALFALFA fluxes are consistent with the conservative source masks put by hand by the ALFALFA team.

The Amount of Diffuse H I
The ratios of diffuse H I over total H I fluxes, f diffuse , are also listed in Table 5.The values of f diffuse range from 5% to 55%, with a median value of 34%.The galaxies with a significant f diffuse (40%) include NGC 628, NGC 5055, NGC 5194, and NGC 5457.These four are the most face-on galaxies in the sample, with inclinations below 60°.The H I disk of NGC 5457 further has the largest R H I in our sample, while NGC 5194 also has a wide-spreading H I distribution due to its ongoing close interaction with the companion galaxy NGC 5195.
The correlations with inclination and disk size reflect a more intrinsic dependence of f diffuse on the minor-axis disk angular size in the H I, as shown in Figure 4 (gray dots). 15By definition, the diffuse H I includes and is slightly more than the missed H I, so any instrumental dependence of f missed as discussed in Section 4.1 may propagate into the diffuse H I. This dependence indicates that the current definition of diffuse H I may be rather dependent on observations (e.g., the physical size, the inclination, the distance of galaxies).In the following, when studying the relation of f diffuse with other galactic properties, we need to consider possible mutual dependence on the inclination or minor axial size.
Despite the possible bias, the diffuse H I in its current definition provides a first-order indicator of the actual diffuse H I that may have a physically clean definition.We can use the current measurements to begin exploring the nature of diffuse H I. In each individual galaxy, it is a relatively consistent measure of the diffuse part of H I in regions throughout the disk and extending to the CGM.We show the spectra of the diffuse H I in Figure 3.The diffuse H I is found throughout the velocity range of each galaxy.Meanwhile, the diffuse H I does not extend beyond the spectral line edges of the dense H I, except for NGC 5194, which is a strongly interacting galaxy.This indicates that most of the diffuse H I is well confined in the potential well of each galaxy, and possibly the bulk motion of the diffuse H I tends to follow that of the dense H I.

Column Density Distributions of the Different Types of H I
We show in panel (a) of Figure 5 the cumulative H I mass distribution in total H I images with column densities greater than the given values.The resolution area over which the column densities are averaged is the FAST FWHM of 3 24, or 9.1 ± 2.5 kpc.It shows that only 5% of the total H I mass is found below a total H I column density of 10 19.8 cm −2 .The fraction is largely consistent with the column density function of Zwaan et al. (2005), indicating that most of the H I masses are found at high column densities even with the image depth reaching 10 17.7 cm −2 .Because this fraction is lower than the median f missed of the sample, while 10 19.8 cm −2 is close to the detection limits of the THINGS data, the missed H I and the diffuse H I are not equivalent to H I found at the low end of the column density distribution.Instead, the missed H I is possibly a large-scale, low-density, and more diffuse part of H I, which is cospatial with the THINGS-detected H I. This implication is confirmed by comparing the conditional cumulative distribution of FEASTS-(denoted by tot) and THINGS-detected H I masses in relation to THINGS-detected H I column densities (blue and purple histograms).All measurements are made at the same FEASTS resolution.These two cumulative distributions flatten below unity when the THINGS-detected column densities are below the detection limits.These flattened parts show the FEASTS-and THINGS-detected H I masses distributed in THINGS-detected regions, which are normalized by the total H I masses throughout the galaxies.The THINGSdetected H I missed the total H I by an average of ∼30%.On average, only approximately one-third of this missed H I is found in regions extending beyond the THINGS-detected regions.Because the missed H I is a subset of the diffuse H I in the THINGS-detected regions, we expect that most diffuse H I is found cospatially with the dense H I.
In panel (b) of Figure 5, we also show the distribution of column densities for the dense and diffuse H I at the same resolution of 3 24.Because the dense H I is not well resolved at this resolution, its column densities are shifted to lower values than their original ones.This plot shows the division in column densities between the dense and diffuse H I when measured at the same resolution.The two types of H I have a rough division at 10 20.1 cm −2 , close to the 2σ threshold estimated with the flux density criteria for selecting the dense H I (Section 3.3).Moreover, the total and diffuse H I have a log N H I dynamic range around four times as wide as the dense H I, hinting significantly more diverse physical conditions traced by them.We first inspect the column density images of the diffuse H I (the left panel in the fourth row of Figures 13-22).In most of the relatively face-on galaxies, there is a smooth distribution of diffuse H I throughout the disks.Around the strong interacting galaxy NGC 5194, there is a significantly enhanced amount of irregularly distributed diffuse H I beyond the disk and tidal tails detected in dense H I. We study this system in detail in another project (X.Lin et al. 2024, in preparation).For moderately interacting galaxies (NGC 5055 and NGC 5457), there is clear evidence of a large amount of diffuse H I linking to the surrounding satellites.The H I bridges have no counterparts in stellar light (Figure 1; also see mid-infrared and violet images in Leroy et al. 2009).The weakly interacting, relatively face-on galaxies (NGC 628 and NGC 925) have neighbors and show lopsided spiral arms, but neither the H I nor optical arms link to their neighbors.In these two systems, the diffuse H I tends to follow the trailing side of the more open arm.The weakly interacting, relatively inclined galaxies (NGC 2841 and NGC 3521) show a tail or warp in diffuse H I on one side, but neither the H I nor optical tails/warps point to the neighbors.These two galaxies may be relatively inclined analogs of NGC 628 or NGC 925.The small neighbors may have closely passed by these four galaxies in the past, inducing the arms, tails, or warps.The remaining highly inclined galaxies seem to have a close neighbor detected in H I, but do not clearly show perturbed features in H I (NGC 2903, NGC 3198, NGC 7331).They also have relatively small amounts of diffuse H I, mostly in the outer disks.
The velocity images of diffuse H I (the middle panel in the fourth row of Figures 13-22) look similar to the velocity images of the total (and dense) H I, indicating that the diffuse H I follows similarly rotating disks as the total (and dense) H I. In some galaxies, there seems to be abnormality in the central regions of velocity fields of the diffuse H I, but we refrain from overinterpreting before a more careful modeling to account for projection and beam-smearing effects is conducted in the future.We focus on the difference velocity images (v dense − v diffuse , the fifth row in Figures 13-22), which show the difference in velocity between dense H I and diffuse H I. For most galaxies, the difference velocity fields show a similar pattern of rotation as the velocity fields of total and dense H I in at least the relatively inner disks.But the amplitudes of such patterns are systematically lower than the latter.This indicates that the diffuse H I in at least the inner disk tends to follow the rotation of the main disk or dense H I, but in a relatively lagged way.The lag of rotation is seen almost throughout the disks of the relatively inclined galaxies NGC 2841, NGC 3198, NGC 3521, and NGC 7331.But it is limited to a relatively small radial range from the center in the relatively face-on and moderately/highly interacting galaxies NGC 5055 and NGC 5457.The velocity difference is less uniform in outer disks, possibly due to the ongoing or past tidal interactions and existence of warps.It is particularly interesting to point out the case of NGC 5457, which seems to have a perfectly lagged Figure 5.The mass and number distributions of column densities for the different types of H I. All measurements are made at the same resolution as the FEASTS (9.1 ± 2.5 kpc).Each galaxy has an equal weight in both panels.(a): Cumulative H I mass distribution as a function of H I column densities.Each set of histograms shows a cumulative H I mass distribution in H I images of type A related by pixels to column densities of type B greater than the given values.The cumulation of a distribution starts from the high column density end and is normalized by the total H I. The A and B types are denoted in the plot, with their images being registered to the same WCS system.The gray dashed line marks a y-axis value of 0.95.(b): Differential H I column density distributions.The blue, red, and magenta histograms are for the dense, total, and diffuse H I, respectively.The areas of histograms are normalized to unity.The dashed line marks a rough division between the two types of H I at a column density of 10 20.1 cm −2 .rotation pattern in the inner disk, but actually the pattern center is off-center with respect to the main disk and dense H I rotation.This feature of shifted center can be seen more clearly in the radial profile of rotational lagging in the next section.
We finally inspect the velocity dispersion images of the diffuse H I (the right panel in the fourth row in Figures 13-22).High-velocity dispersions of the diffuse H I seem a ubiquitous feature observed throughout the sample.In relatively face-on galaxies, velocity dispersions of the diffuse H I are typically higher than the dense H I throughout the disks.Their inner disks have higher velocity dispersions in the diffuse H I than the outer disks.In the inner disks of highly inclined galaxies, we often see X-shaped structures, which are possibly caused by projection and beam smearing.

Radial Distribution of Diffuse H I Properties
We present the radial distribution of diffuse H I properties, including the column density, extent of lagged rotation, velocity dispersion, and kinematical hotness (i.e., the extents of motion being dominated by random instead of rotational motions), in Figure 6.We compare the diffuse H I to dense H I in most properties, which are measured at the same FAST resolution of 3 24.The plots summarize our visual impression on individual moment images in a more quantitive and statistical way.We caution that all the results below are based on relatively coarse and first-order measures, and we will need careful kinematical modeling to better account for possible beam smearing and projection in the future (T.Randriamampandry et al. 2024, in preparation).Meanwhile, all profiles start from the radius equal to the FWHM of the FAST beam, while the extended nature of the diffuse H I makes it less susceptible to beam smearing or smoothing than the dense H I. The beam effects are further mitigated when we focus on the relative ratio instead of absolute differences between the diffuse and dense H I.  In panel (a) of Figure 6, we show the radial profiles of diffuse H I column densities.Most of the diffuse H I has column densities below the threshold of the dense H I. They generally tend to flatten near the center and decrease with radius, similar to radial profile shapes of normal H I column densities (Wang et al. 2016).They do not exhibit obviously abrupt truncation at a large radius, implying a smooth transition to the CGM at least down to a column density limit of 10 18 cm −2 , though future modeling will be needed to fully rule out spatial smoothing effects of the beam.
The diffuse H I extends out to 4-7 times the optical radius.We obtain from the radial profiles the characteristic radius r 18 , where the column density of diffuse H I reaches 10 18 cm −2 .We plot in Figure 7 the r 18 as a function of the B band luminosity.We compare such a relation to that of the fiducial boundary radius (R gas ) for Mg II-traced cool gas detected in the CGM, which is Chen et al. 2010), where L B is the B band luminosity (from W08), and L B * is the characteristic luminosity of the luminosity function in the B band from Faber et al. (2007).The r 18 values are systematically lower than, and thus consistent with, the cool gas boundary radius R gas at a given B band luminosity.The ratios r 18 /R gas have an average value of 0.74 ± 0.15.This indicates that, around H I-rich galaxies like those in the THINGS sample, the H I or cool gas detected in the CGM is possibly linked to the disk instead of being floating clouds.
In panel (b) of Figure 6, we show the radial distribution of the ratio of diffuse H I over the total H I. The diffuse H I contributes at a low level to the total H I within R 25 in most galaxies, except for the most strongly interacting galaxy NGC 5194.For the remaining less interacting galaxies, the contribution of diffuse H I to the total H I increases toward large radii, roughly with an exponential profile.For the three galaxies with the lowest f diffuse (NGC 2841, NGC 3198, NGC 3521), the ratio of diffuse H I rises above half only when r > 4 R 25 .This confirms the more extended nature of the diffuse H I. A detailed comparison in radial profile shapes between the total, dense, and diffuse H I will be presented in a future paper.
In panel (c) of Figure 6, we show the velocity difference between the dense H I and diffuse H I (v r,dens − v r,diff ) divided by sin i for both the receding (solid curves) and approaching (dashed curves) sides of galaxies.The sin i term uses the optical inclination angle and corrects the radial velocities for projection assuming rotational motions.We can see that, for most galaxies, on the receding (approaching) sides of disks, the velocity differences tend to be positive (negative).Because both lagged rotation and counterrotation lead to positive (negative) velocity differences on the receding (approaching) sides, it can be more informative to check the normalized velocity difference between the dense H I and diffuse H I ((v r,dens − v r,diff )/v r,dens ; see Section 3.3.1).
In panel (d), we can see that the normalized velocity difference profiles are largely positive and within unity for both the receding and approaching disk sides of most galaxies.The behavior of most galaxies indicates that the majority of diffuse H I in the sample is rotating in the same direction but in a lagged way in comparison to the dense H I. The rotations of the diffuse H I lag behind the dense H I by a median of 25% in velocity in the radial range of 1-2 R 25 , but by as high as ∼50% within R 25 .On the whole, the extents of lagging drop from a small to a large radius.This lagging is unlikely due to beam smearing: because the diffuse H I is less (centrally and locally) concentrated than the dense H I, its observed velocity is also less artificially "lagged" by beam smearing.Because the dense H I is possibly more strongly smeared and artificially "lagged" when smoothed to the FEASTS resolution, the actual extents of lagging of the diffuse H I may be even higher than shown here.On the outskirts (r > 1.5 R 25 ), some profiles become negative possibly due to tidal perturbations (i.e., the gas rotates faster or deviates from the circular orbits).NGC 5194 is an outlier in this plot, with normalized velocity difference on the receding side rising as a function of radius from the center, and reaching values above unity; its profile for the approaching side is not shown due to values larger than unity, implying orbiting in the opposite direction with respect to the dense H I.
Panel (e) of Figure 6 shows the velocity dispersion radial profiles for diffuse H I. These profiles drop from the galaxy center toward outer disks, a trend that was seen for the dense H I (Tamburro et al. 2009), and was interpreted as a result of energy input from stellar feedback and gas inflow activities.The velocity dispersions of diffuse H I have values roughly twice those of dense H I (dotted curves), the latter showing values close to 10 km s −1 typically observed for high-column density H I (Tamburro et al. 2009).
Panel (f) shows the kinematical hotness of the diffuse H I disk, measured as the ratio of velocity dispersion over the rotational velocity approximated as the projection-corrected radial velocity, Similar results based on analysis of the missed H I are presented in Figure 23 in Appendix G.The plots show that distribution and kinematical properties of the missed H I are largely consistent with those of the diffuse H I. But we highlight that, from panel (a) of Figure 23, most of the missed H I has column densities below the detection limit of THINGS.This implies that most of the THINGS missed H I is due to its low density and not necessarily large angular scale.This is different from the missed H I in W23 comparing the FEASTS and HALOGAS data, where 90% of the diffuse H I is missed by HALOGAS due to its large angular scale instead of low density.

Discussion
In this section, we discuss possible physical scenarios related to the properties observed for the diffuse H I.

Kinematically Warm Disks of Diffuse H I
Keeping in mind that the projection and beam-smearing effects are not perfectly accounted for, we find that, for most galaxies, the diffuse H I is organized in rotating disks that resemble but are kinematically warmer than the dense H I disks.The kinematics of the diffuse Hi is largely rotation-dominated, and tends to follow the rotating direction of the dense H I throughout the disks.But its rotation is usually slower than the dense H I by around 25%, and its velocity dispersions (typically 10-25 km s −1 ) are twice that of the dense H I. These kinematical features are unlikely dominantly caused by beamsmearing effects, as the diffuse H I distributes in a less concentrated way than the dense H I. The "lag in rotation" is closely related to the disk asymmetric drift, suggesting kinematics contributed partly by random instead of purely rotational motions.But it can further involve dynamic mixing with the CGM, and possible inflow motions due to cooling.The diffuse H I is thus likely a kinematically hotter extension of the dense H I into the hotter and more ionized regime of the CGM.Because the diffuse H I dominates the total H I at a large radius, while the dense H I is usually related to star formation at a small radius, we separately discuss the possible scenarios in relatively inner and outer disks, roughly divided by 1.5 R 25 .

The Similarity of Diffuse and Thick H I Disks
The kinematics of diffuse H I disks except for the very outskirts in a few galaxies resembles that of the thick H I disk identified in deep interferometric H I observations (e.g., Oosterloo et al. 2007;Heald et al. 2011), as they both show lagged rotation and high-velocity dispersions within roughly the optical radius (Marasco et al. 2019).Despite different technical details, there is no obvious reason to consider the two types of disks as being intrinsically different.
Galactic fountains, which have been suggested to be an important channel to produce the thick H I disks (Fraternali & Binney 2006), should also contribute to producing the diffuse H I detected in this work.Future efforts on kinematical modeling of diffuse H I, revealing multiphase gaseous structures using multiwavelength data, and linking them to the localized SFRs, will be needed to quantitatively confirm the role of fountains.For now, if we take observations of the high velocity clouds consisted thick H I disk of the MW (the properties of which are highly consistent with being fountain driven) as the reference, the lagged extents of rotation at the level of ∼20% and 10% at a radius around 1.5 and 2 R 25 in our sample correspond to a height of 3 and 1 kpc above the disk plane (Marasco et al. 2012), respectively.

The Connection of Diffuse H I to CGM Properties in the Outer Regions
We perform an order-of-magnitude estimation on the thickness of the diffuse H I in outer disks, largely following the method described in W23.We assume the diffuse H I to be internally supported by turbulent pressure (mass per particle μ = 1.273 m H ), and externally in pressure equilibrium with the hot CGM (μ = 0.609 m H ). We take the group mass of each galaxy from Kourkchi & Tully (2017) and estimate the viral temperature.We estimate the CGM gas mass from the group mass using the equation 5 of Ettori (2015), and assume the CGM gas distribution following an isothermal β-model with β = 0.64 and core radius equal to 0.19 r 500 , where r 500 is the radius of the group where the average density is 500 times the critical density of the Universe.The estimated CGM density radial profile is present in panel (a) of Figure 8.We divide the column density of diffuse H I by the volume density and obtain the thickness.Because the diffuse H I seems to largely follow the rotation/orbits of the dense H I, the column densities have disk-like projection effects, and the derived thickness should be viewed as upper limits.We also multiply the column density of diffuse H I by cos i to obtain the "projection-corrected" column density before deriving the thickness, where i is the inclination of the dense H I disk (Table 1).This projection correction assumes the H I to be in a thin disk, and the corrected column density and thickness should be lower limits.
As we show in panel (b) of Figure 8, the thickness profiles of all galaxies decrease with radius.Except for in NGC 925, the directly derived thicknesses of diffuse H I of all galaxies are less than 3 (1.5)kpc when the radii are larger than 1 (1.5)R 25 . 16 The projection-corrected diffuse H I thicknesses of all galaxies are below 3 kpc outside R 25 .The real thicknesses are possibly between the projection-corrected and uncorrected values.The typical thickness of diffuse H I in outer disk regions in this study is thus much smaller than that inferred in a similar way for NGC 4631 in W23, which is 4.5 and 10.0 kpc at a galactocentric distance of 30 and 55 kpc.The reason is that most column densities and velocity dispersions of diffuse H I derived in this study are slightly and much lower than the typical values of diffuse H I detected around NGC 4631 (10 19.7 cm −2 and 50 km s −1 , respectively).
Figure 7.The relation between r 18 and normalized B band luminosity.r 18 is the characteristic radii where the azimuthally averaged column density of diffuse H I reaches 10 18 cm −2 in a galaxy.The curve shows the relation between R gas and the normalized B band luminosity from Chen et al. (2010), where R gas is the typical border radius for cool gas to be found in the CGM. 16These thickness values can increase by roughly 2 times if the localized ionizing rate is above 0.
In panel (c) of Figure 8, we plot the relation of thickness against column density for the diffuse H I in each galaxy.The different intercepts of relations probably reflect projection effects, but it is surprising that each relation is nearly linear, regardless of applying projection corrections or not.Such linear trends imply a constant volume density for the diffuse H I in each individual galaxy, while the CGM volume densities vary for more than 0.5 dex in the same radial range.Because the gas cooling rate is correlated with the gas volume density, the diffuse H I possibly marks a shell of a roughly constant cooling rate.This implies that the diffuse H I detected here may indeed be a cooling interface between the ISM and CGM.We emphasize that the thickness values derived in this way have large uncertainties due to the simplified assumptions of the CGM model and pressure equilibrium status, and beam smoothing/smearing and projection effects.These estimates should be revisited in the future when the diffuse H I distribution and kinematics are better modeled and direct measurements of CGM densities and temperatures are available.
To sum up, while the diffuse H I detected here has spatial scales (18 kpc, Section 3.2) similar to that of NGC 4631, its thickness is much smaller.It is more layer like, while the latter is more halo like.Nevertheless, they may both have close relations to CGM cooling.The constant volume density beyond R 25 tentatively supports its role as a cooling interface between the ISM and the CGM for the galaxies in this study, while the 10 19.7 cm −2 level high-column density of the diffuse H I in NGC 4631 indicates an efficient cooling flow driven by thermal conductions.

The Possibly Important Role of Tidal Interactions in
Producing Diffuse H I Keeping in mind that the sample is relatively small, we find that the fraction of diffuse H I ( f diffuse ) seems to be related to the morphology and kinematical abnormalities in the diffuse H I, which implies intensities of tidal interaction (Section 4.4).We summarize this trend in Figure 9.Such a relation between diffuse H I and tidal interactions has been speculated in W23 based on a detailed study of the interacting system NGC 4631, but here studied in a statistical way.We also include NGC 4631 in this figure.The fraction and column density maps of the diffuse H I in NGC 4631 have been derived using the method of this study, but they both should be taken as lower limits, for the HALOGAS data have better sensitivity on short baselines than the THINGS data.
As in the top panel of Figure 9, the interaction intensities can be roughly ordered by the H I morphologies as having no significant H I tidal features, having H I tails or warps possibly induced by a past encounter, having a tidal bridge linking to a small companion galaxy, and being in the matured stage of a major merger.The highest level of interacting intensity is further supported by kinematic abnormalities in difference velocity fields in the outer disks (NGC 5194 and NGC 5457; Section 4.4.1).This relation is likely robust against the systematic dependence of f diffuse on galaxy minor-axis angular sizes, because it exists among both relatively face-on and inclined galaxies.Moreover, the diffuse H I becomes more dominating in the total H I along tidal features typically found at a large galactocentric radius, which also supports its link to tidal interactions.The bottom panel of Figure 9 shows the corresponding morphologies in the optical, but most tidal features in the H I are not seen in the optical.In previous observational studies, it is common for interferometers to detect extraplanar H I (de Blok et al. 2014) and single-dish telescopes to detect diffuse H I in tidal interactions (Borthakur et al. 2010).The main progress here is that the diffuse H I is mapped to a relatively low column density limit at a moderate resolution for a relatively large sample (thanks to the large diameter and Figure 9.The variation of galaxy H I morphology in the space of tidal interaction strength and diffuse H I fraction.Top: the stamps are moment 0 images of the diffuse H I. The galaxy names are denoted.The fraction and column density maps of the diffuse H I in galaxy NGC 4631 have been re-derived using the method of this study, but the f diffuse of NGC 4631 should be taken as a lower limit, because its HALOGAS data have better sensitivity on short baselines than the THINGS data.Bottom: the stamps are contours of total H I column density levels overlaid on top of the optical images.The outmost contour starts from the detection limit and is always dominated by the diffuse H I. The outer regions of the total H I are always dominated by the diffuse H I, and because the diffuse H I distribution is quite flat, its contours will largely be the outmost contours of the total H I. 19-beam receiver of FAST), so that the distribution and kinematics of diffuse H I directly show tidal features.
Besides stellar feedback (fountains) strengthened due to tidally induced star formation in gas-rich galaxies (Moreno et al. 2015), theoretical studies predict that tidal interactions may produce the diffuse H I in at least the following three ways.First, tidal interactions strip H I initially in disks into tails and bridges, and enhance turbulent mixing that may produce the diffuse H I. Simulations predict that, when neutral gas moves subsonically in the CGM, a cooling-dominating turbulent mixing layer develops with the column density correlated with the initial traveling speed and the velocity dispersion one or two magnitudes lower than the initial traveling speed (Ji et al. 2019;Yang & Ji 2023).The velocities of tidal tails and bridges in our sample rarely exceed the maximum rotational velocity of the dark matter halo, and thus should be mostly subsonic; the velocity dispersions of the diffuse H I are on the level of 20 km s −1 , consistent with predictions for gas in the turbulent mixing layer.The diffuse H I near the tidal bridges and tails of NGC 5194, NGC 5457, and NGC 5055 may be related to the turbulent mixing.Second, tidal interactions drive spiral arms in galaxies and produce the diffuse H I through shocks of gas running into arms.Possible candidates in support of this scenario include NGC 628 and NGC 925.They have lopsided spiral arms, and the diffuse H I tends to follow the trailing side of the more open arm, which is likely rotating faster toward a large radius than the other arm.Spiral arms in NGC 628 and NGC 925 have a pattern speed of 41.8 and 7.7 km s −1 kpc −1 , respectively, derived in the literature (Elmegreen et al. 1998;Martínez-García & Puerari 2014).It only takes a time lag of 66 and 12 Myr, respectively, to fill an angle of 30°behind the arms and produce the observed morphology in these two galaxies.Third, interaction with companion galaxies may directly heat the H I disk, with gravitational and hydrodynamic effects.This mechanism may contribute to the diffuse H I in the inner disks of NGC 628, NGC 925, NGC 2841, and NGC 3521.

The Abnormal Diffuse H I in NGC 4631
A question after comparing the sample studied here with NGC 4631 (W23) is why does NGC 4631 stand out?The HALOGAS data are an order of magnitude deeper in column density than the THINGS data, and their uv coverage on short baselines by design is much denser than the latter, yet they missed 40% (>70%) of the H I fluxes detected by FEASTS in the NGC 4631 system (CGM region), the central galaxy of which is edge-on (W23).In order to clearly demonstrate this contrast with the THINGS sample, we conduct the same analysis as in this study for the THINGS data to the HALOGAS data of three overlapping galaxies, NGC 925, NGC 3198, and NGC 5055.The derived fraction of missed H I from the HALOGAS images is 12%, 0%, and 8% for these three galaxies, respectively, in comparison to 24%, 0%, and 26% from the THINGS images.These HALOGAS-missed H I fractions decrease to 10%, 0%, and 4% when the WSRT primary beam is applied to both HALOGAS and FEASTS images (i.e., when the contribution from primary beam attenuation is removed).These comparisons suggest that the HALOGAS images have smaller short spacing and sensitivity limitations than the THINGS data in detecting the diffuse H I. NGC 4631 is indeed a special system for which even the HALOGAS observation misses a large fraction of H I in comparison to the FEASTS data.If it were observed by THINGS, the missed H I fraction would have been even higher, possibly higher than all the THINGS galaxies studied in this paper.
The diffuse H I of NGC 4631 has distinct column densities (3 times higher), velocity dispersions (∼2.5 times higher), and inferred thicknesses (∼10 times higher) in comparison to all other galaxies studied here.Most of the diffuse H I is distributed perpendicular to instead of along the disk of NGC 4631, again in contrast to the sample here.NGC 4631 is indeed an extreme case in terms of its halo properties, not only in diffuse H I, but also in radio continuum, Hα, X-ray, etc. (Martin & Kern 2001;Wang et al. 2001;Irwin et al. 2012).All these may be related to the extensive tidal tails.It is then particularly interesting to compare between NGC 4631 and NGC 5194, both of which have grand-design tidal structures in the dense H I and particularly in the diffuse H I.
Do these two galaxies have different degrees of galaxygalaxy interactions?The strength of galaxy-galaxy interactions depends on the mass ratio, orbits, number of involved galaxies, times of encounters, etc. NGC 4631 is 2.5 times less massive than NGC 5194, but its major companion NGC 4656 is 16 times less massive than NGC 5195, which interacts with NGC 5194.The lower mass ratio in the NGC 4631 system, which implies a lower degree of tidal interaction, is unlikely to explain its more extreme diffuse H I. But NGC 4631 is simultaneously interacting with four H I-rich neighbors (Rand 1994), and strongly with at least two of them (Combes 1978), while NGC 5194 seems to be mainly interacting with NGC 5195 (Dobbs et al. 2010).The interactions with NGC 4631 mainly occur in the polar direction (Combes 1978), while those with NGC 5194 are rather coplanar (Dobbs et al. 2010).Hydrodynamic simulations demonstrate that highly inclined orbits sometimes lead to stronger perturbations than coplanar orbits (Bustamante et al. 2018).It is possible that the large numbers of involved galaxies and the polar orbits result in higher degrees of tidal interactions, leading to more efficient diffuse H I production around NGC 4631.
The stellar feedback may also play a role.Though the two galaxies have similar SFRs (4.2 versus 4.4 M e yr −1 ), NGC 4631 has a 2.5 times lower stellar mass than NGC 5194, and its SFR distributes in more clustered ways, producing the super shells observed in H I and molecular gas (Rand 1994(Rand , 2000)).Such concentrated, bursty star formation in a lower gravitational potential produces more effective stellar feedback.
The intensive stellar feedback and extensive tidal interactions in NGC 4631 may produce its abnormal, halo-like diffuse H I. These two effects may together drive large-scale CGM turbulence followed by a top-down cascade in the multiphase gas (Gaspari et al. 2018); warm ionized and neutral gas condense out of the turbulent eddies, producing diffuse H I half-way between the two.The difference from the turbulent mixing at the neutral gas-CGM interface is that it is a topdown condensation starting from the large-scale CGM.One interesting prediction of this scenario is that the ensemble velocity dispersions of H I on scales of tens of kiloparsecs trace the cascading turbulent velocity of the CGM where the H I condenses out (Gaspari et al. 2018).The turbulent diffuse H I of NGC 4631, which has a velocity dispersion of 50 km s −1 in an ensemble resolution element of 8 kpc (W23), can be a promising candidate to support this theoretical scenario.

Implications on Gas Accretion
The diffuse H I may trace a transition between the dense H I and the CGM, possibly serving as an interface of gas accretion from the CGM.Theoretically, channels for gas accretion onto massive nearby galaxies can be roughly divided into cosmological hot accretion, fountains, and mergers (Putman 2017;Grand et al. 2019).In practice, the first channel is hard to directly observe, but is likely mixed with the latter two channels (Grand et al. 2019).Through comparing the SFR and mass and kinematics of the thick disk H I and associated warm/ hot ionized gas, circulation of gas through fountains seems to be an important gas accretion channel in the nearby starforming galaxies, including the MW (Fraternali 2017;Richter 2017;Marasco et al. 2022).Since the diffuse H I in the inner disk region resembles the thick disk, it may contribute to gas accretion through the fountain channel.The importance of the last channel, tidal interactions, or mergers, seems observable but mostly inconclusive.
First of all, whether mergers lead to an increase of H I richness (at a given stellar mass) has been inconclusive.Theories and simulations suggest its feasibility (Gaspari et al. 2018;Sparre et al. 2022), while recent observations suggest that the H I gas and total neutral gas masses in post-mergers and merger-induced luminous infrared galaxies (LIRGs) seem to be similar to normal star-forming galaxies (Ellison et al. 2018;Shangguan et al. 2019).The cooling of hot gas should not immediately lead to an increase in the dust content of a galaxy, so one may not use dust emission to trace an increase in total neutral gas content.On the other hand, the conversion of H I to molecular gas possibly becomes faster and more important in LIRGs (Bellocchi et al. 2022), which limits the use of total H I measurements.The nearby galaxies of more normal star formation may be a better laboratory to study the physical process of tidally induced gas condensation, and the diffuse H I may provide a unique tracer.
Second, whether merger-related gas accretion is important for general star-forming galaxies has been questionable.Through counting the H I masses in satellite galaxies and deriving the merger timescale, it was largely ruled out that direct merger or stripping of low-mass H I-rich satellites can sustain the star formation in central galaxies (Kauffmann et al. 2010;Di Teodoro & Fraternali 2014).However, the ubiquitous existence of diffuse H I around H I-rich galaxies and the link to tidal interaction strengths, together with the theoretical possibility of large-scale gas condensation due to turbulence cascade and small-scale cooling flow due to turbulent mixing, seem to imply that tidal interaction with neighbor galaxies may bring more cool gas into a galactic disk than just adding the neutral ISM of satellites.As almost all galaxies in our sample have H I-bearing small companions, wet tidal interactions of nearby H I rich galaxies can be more frequent than reported in optical flux limited observations or in resolution limited simulations.The role of tidal interactions in gas accretion and star formation fueling may thus be worth reconsideration.
Nevertheless, linking diffuse H I to cooling in the CGM is a complicated problem, because the diffuse H I is only an intermediate phase.The fate of diffuse H I depends on the interplay between many physical processes including UV background ionization, Rayleigh-Taylor instabilities, Kelvin-Helmholtz instabilities, turbulent mixing and cascading, radiative cooling, and thermal conduction (Tumlinson et al. 2017).These processes can be better characterized by observationally dissecting the multiphase gas, as well as obtaining information on the metallicity, magnetic fields, and cosmic rays.We hope our statistical characterization of diffuse H I may provide motivation for observational efforts on multiphase gas and useful constraints on theoretical efforts.

Summary and Conclusion
We have analyzed in combination the FAST FEASTS images and VLA THINGS images of H I for 10 nearby galaxies.Special care has been taken to ensure flux density scale alignment between the two observations, so that we can robustly quantify the excess H I detected by FEASTS in comparison with the THINGS data.Missing H I from the THINGS data is robustly detected in at least 8 out of 10 galaxies.The integral H I mass from interferometric observations of nearby large galaxies should thus be used with caution.
Diffuse H I is defined as the difference between total and dense H I, while the dense H I is selected from the THINGS cubes with a relatively uniform flux density threshold.The fractions of diffuse H I over the total H I range from 5% to more than half with a median value of ∼34% among the sample.From analyzing the distribution, velocity, and velocity dispersion of the diffuse H I, we find that most of the diffuse H I is likely organized in a rotating disk that is kinematically hotter than the dense H I. The diffuse H I detected here is more like a thin layer with around 1 kpc thickness, either by itself in the outer galactic region or as a shielding layer at a height of 1 to 3 kpc above the thin disk of dense H I in the inner region, instead of halo-like structures detected in NGC 4631.Thus, the diffuse H I may have different categories and possibly different producing mechanisms.Tidal interaction intensities seem to be closely linked to the fraction, morphology, and kinematics of diffuse H I detected in galaxies.While fountains may contribute more to the inner star-forming disks, tidal interactions may be an efficient producer of diffuse H I throughout the disks of general H I-rich galaxies.
The possible role of diffuse H I as an intermediate phase between dense H I and the CGM, and the theoretical links of fountains and tidal interactions to gas condensation, indicates a promising role of diffuse H I in dealing with the gas accretion problem.Separating these different mechanisms, and better understanding the fate and role of different types of diffuse H I in galaxy evolution, will be major tasks in upcoming work, as more data (for a planned sample of 118 galaxies in total) are obtained and analyzed in the FEASTS program.
images and obtain the single-dish FFT image and the interferometric FFT image, respectively.We select the FFT pixels that are in an overlapping frequency range with the lower and upper limits determined by the largest and smallest spatial scales detected in the interferometric and single-dish data, respectively.We further select the pixels that have a S/N above 4 in both FFT images.The cross-calibration factor of fluxes f F/T is derived as the 3σ clipped mean of the ratio between the single-dish FFT amplitudes and the interferometric FFT amplitudes of the selected pixels.

B.2. Updated Beam Shape and Beam Area
Compared to W23, we have updated the average beam image of FAST.The average beam shape when the receiver has a position angle of zero has been used in W23, but in FEASTS observations, the receiver was rotated to achieve the Nyquist-Shannon sampling rate.The unrotated receiver of FAST has an m = 6 harmonic pattern in the outer region of the beam image.As the rotated receiver of the horizontal and vertical scans have position angles differing by 30°, the m = 6 harmonic patterns of the two rotated beams cancel out.As a result, the effective average beam of FEASTS has an almost azimuthally symmetric shape in the outer region, as shown in Figure 10.
The radial profile of the beam does not change in comparison to the one showed in W23.We also point out that, as most of the diffuse H I in W23 has high column densities, this correction in beam shape does not significantly affect the results on diffuse H I in W23.
As in W23, the beam FWHM of 3 24 is confirmed by fitting double Gaussian models to the beam image.But the actual beam area is larger that of the Gaussian beam by a factor of 1.064 due to the existence of side lobes.This factor is considered in all cases when converting the unit of Jy beam −1 to Jy arcsec −2 or to Jy pixel −1 .

B.3. Adjustment in the Cross-calibrating Procedure
We adjust and improve the procedure to optimize it for the THINGS data that are shallower than the HALOGAS data used in W23.Many of these adjustments are developed based on tests with mock data that have similar qualities as the FEASTS and THINGS data.The mock simulation and the tests are presented in detail in Appendix C. The adjustments are summarized below.
1. We use the moment 0 image (produced in Sections 2.2 and 2.3) instead of the cube to conduct the comparison.2. The THINGS moment 0 image used is derived from the rescaled cube, using the SoFiA mask of the standard cube.Using this type of moment 0 image to correctly recover as much faint and extended fluxes as possible is important in order to have a consistent comparison in amplitudes with the FEASTS data.We show that the rescaled cubes are much better than the standard cubes, and slightly better than the convolved model cubes for cross-calibration based on tests with mock data in Appendix C. 3.Only the inner region with a radius of 750″ is used for the calibration, as the primary beam of THINGS data quickly drops below 0.5 beyond that.4. The upper limit of spatial frequency is set to be 1.5 times the FAST beam size as before, while the lower limit of the spatial frequency is set to be 8 82, corresponding to a baseline of 100 m.Like in W23, the interferometric data start to miss fluxes in comparison to the FAST data before reaching the largest scale allowed by its shortest baseline.The reason is likely a combination of limited sampling density of the shortest baseline, which can be worsened by RFI flagging (like in W23) and the relatively high rms level due to limited integration time (unlike the case of W23).

B.3.1. Arguments for Using Moment 0 Images in Cross-calibration
Among the many adjustments listed above, a most controversial one may be using moment 0 images instead of the more conventional choice of cubes or visibilities for crosscalibration.The data cubes already have the shortage of having nonlinear noises introduced by the CLEAN, while the moment 0 images further have the disadvantage of being after noise dependent thresholding of source finding.
A first support of this method can be based on the assumption that interferometric moment 0 images widely used for characterizing H I distributions should be reasonably good in regimes where the S/N and uv sampling are sufficiently good.But, more importantly, the data quality requires us to use the moment 0 images instead of cubes for cross-calibrating.First, a sufficiently high S/N level is required for a robust cross-calibration.We show in Appendix C that the interferometric image should have a median S/N above 2.6, in order for the systematic uncertainty of cross-calibration to be below 5%.Most THINGS moment 0 images have a median S/N just above this threshold (Table 3), while the related channel maps only have S/Ns roughly one tenth that level.Increasing the S/N through a sufficient smoothing in the spectral direction is almost equivalent to making the moment 0 image.Smoothing in the image domain does not help either, as it is already included in the source-finding (smoothed with a Gaussian kernel that has a FWHM of 30″) and cross-calibrating (smoothed with the FAST beam) procedures.Second, a sufficiently wide dynamic range in the uv coverage, or spatial scales of H I structures, is necessary for a robust cross-Figure 10.The average beam image accounting for the rotated receivers.The white, dotted contours are at levels of 5e-4, 5e-3, 5e-2, and 0.5.The levels are also marked in the color bar.A solid white circle is plotted near the outmost contour for reference, to highlight that the averaged beam is circular in the outer region.
Figure 11.The cross-calibration procedure plots for each galaxy.Left: the FFT amplitudes of the FAST and VLA moment 0 images as a function of angular scales (the inverse of spatial frequency).The FAST and VLA measurements are plotted in red and blue, respectively.The two vertical lines mark the lower and upper limits of the overlapping angular scales, between which the pixels are selected.Middle: comparing the FFT amplitudes between the two types of data.The FFT amplitudes of selected pixels from the two types of FFT images are plotted in gray dots, and the gray dashed line has an intercept of 0 and a slope of the derived f F/T .The imaginary/ real parts of selected pixels from the two types of FFT images are plotted as orange and purple dots to confirm the goodness of the derived f F/T .The black solid line shows the position of y = x.Right: the ratio of corrected FAST amplitudes over VLA amplitudes as a function of angular scales.The two vertical lines from the left panel are repeated.The dashed horizontal line marks the position of y = 0. calibration (Stanimirovic 2002;Kurono et al. 2009).In principle, the lower and upper limits of the overlapping region are determined by the FAST resolution and the shortest baseline of the THINGS observation, but the real flux structures do not always cover the selected region with sufficient power, which is exacerbated by the limited sensitivity.In many channel maps of the THINGS cube, the flux-detected regions can be smaller than the FAST beam.This problem is much mitigated in the moment 0 image.
A possibly most fundamental and accurate (but see Stanimirovic 2002) but computational expensive method is to convert the FEASTS images to visibilities and compare the visibilities for cross-calibration.This method has advantages of keeping the interferometric data closer to observations, being less affected by filtering and thresholding.However, this method should also be limited by the noise level of data and spatial spanning of fluxes.Moreover, producing simulated visibilities for single-dish images introduces interpolation errors.These effects remain to be explored.
Thus, in this first study of combining FEASTS data with interferometry for a relatively large sample of galaxies, we adopt and optimize the relatively simple and quick method of cross-calibrating with moment 0 images.This method can be easily used by nonradio-experts and can be applied to combining large samples of interferometric images and single-dish images.Such images are likely to be available in the near future with WALLABY (Koribalski et al. 2020), Apertif (Adams et al. 2022), and CRAFTS (Zhang et al. 2019) wide-field H I surveys.

B.4. Results of Cross-calibration
Figure 11 presents the results from the flux cross-calibration procedure.Each row is for a target galaxy.The left panel of each row shows the FFT amplitudes of the FAST and THINGS moment 0 images as a function of angular scales.The two vertical lines mark the lower and upper limits of the overlapping angular scales, between which the pixels are selected.We can see that between the two vertical lines the two types of amplitudes often have a similar range of values, but beyond the vertical line of the upper limit the FAST amplitudes are often higher.In the middle panel of each row, the gray dots show the relation of FFT amplitudes from selected pixels between the types of data, and the gray dashed line has an intercept of 0 and a slope of the derived f F/T .The gray dashed line is usually not far away from the black line of unity.We also plot the imaginary/ real parts of selected pixels from the two types of FFT images.Most of the data points follow the relation of the amplitudes, justifying the robustness of the cross-calibration.In the right panel of each row, we plot as a function of angular scales the ratio of corrected FAST FFT amplitudes over VLA FFT amplitudes.The scaled FAST FFT amplitudes are now consistent with the VLA FFT amplitudes within the selected angular range, but start to be higher than the VLA amplitudes roughly when the angular scale is larger than the upper limit, which is roughly 8 82, or 18.7 to 37.7 kpc in this sample.

Appendix C Mock Test to Optimize the Flux Cross-calibrating Procedure
The primary questions to address here are whether we can achieve reasonable accuracy in flux cross-calibration between the VLA and the FAST data, and what type of VLA image is best for the cross-calibration.The candidates of VLA maps include the standard clean map, the rescaled clean map, and the convolved model map.To address these questions, we simulate mock images of disks, which have size, power-spectrum slope, and S/N close to the moment 0 images of real galaxies.We analyze them in the same way as we do for real data, and check how the derived scaling f F/T deviates from the actual value of unity.

C.1. Mock H I Disks
We describe below the step of simulating a mock H I disk.We first create an image of a face-on disk that has an R HI of 9′ and a surface density radial profile following the median profile of Wang et al. (2016).The H I disk is larger than most target galaxies in our sample, so the mocks represent the most difficult case for VLA observations.The image size is set to be 2000 pixels on each size, and the pixel size 1 5.We use the Python package TurbuStat (Koch et al. 2019) to create a synthetic image of the same size based on a power-law noise distribution.The power-spectrum slope is set to a value −k close to real galaxies in H I moment 0 images (see Appendix D).The flux scale of the synthetic noise image is globally normalized to have values between −1 and 1.The synthetic noise image multiplying with the face-on disk image is added to the face-on disk image.Any pixels below zero are set to be an arbitrarily positively low value.Finally, the pixel values are globally scaled to have a total flux of 23.22 Jy.We have produced mock H I disks that have similar small and large-scale structures as real H I disks.We produce mock H I disks with six different k values, ranging from 1.8 to 3.2.The peak fluxes of these disks range from 0.3 to 0.14 mJy pixel −1 .

C.2. Mock FAST Images
The mock H I disk is convolved with the FAST beam image, which has been resampled to have the same pixel size.We generate a noise image following a random Gaussian distribution with a σ width of 0.57 mJy beam −1 .The σ looks lower than the values listed in Table 2 because the FAST data are smoothed along the velocity direction when reprojected to the WCS system of the VLA data.We add the noise image to the mock disk convolved with the FAST beam, and this produces a mock FAST image.

C.3. Mock VLA Images
We use the CASA task simobserve to simulate VLA observations and produce mock visibilities.We feed the mock disk to the task as the sky model.We use the coordinate of the galaxy NGC 628 as the direction of pointing.To mimic the uv coverage of real THINGS observations, we simulate three sets of observations in the B, C, and D array configurations, with integration times of 7, 2.5, and 1.5 hr, respectively.We follow the lengths of integration time of the THINGS observation of NGC 628 to achieve similar uv coverage.We only simulate one channel map, and thus can conveniently adjust the rms level by changing the channel width of the observation.We use channel widths of 0.25, 0.05, 0.04, 0.03, 0.02, 0.01, and 0.001 MHz to achieve an rms level of 0.31, 0.50, 0.53, 0.60, 0.71, 0.96, and 2.86 mJy beam −1 .We note that the S/N ratio distribution in the whole flux-detected region is more informative than the absolute rms level in the problems we discuss.Later, we find

Figure 1 .
Figure 1.False-color images of FEASTS-detected H I in the optical image of galaxies.The field of view and coordinates of these plots are similar to those in Figures 13-22 in Appendix F. The small circle and horizontal bar shown at the bottom-left corner of each panel show the beam size of the FEASTS observation and a length of 10 kpc.The optical images are from the Legacy Survey (Dey et al. 2019).

Figure 3 .
Figure3.The integrated H I spectra from FEASTS and THINGS data.The red solid lines and blue dashed curves are for the FEASTS and THINGS spectra, respectively.The cyan dotted curves plot the spectra of the missed H I, which is the FEASTS spectrum minus the THINGS spectrum for each galaxy.The magenta dotted curves plot the spectra of the diffuse H I, which is the FEASTS spectrum minus the dense H I spectrum for each galaxy.

Figure 4 .
Figure 4.The change of missed or diffuse H I fraction as a function of angular sizes of galactic H I minor axes.The diffuse H I fractions f diffuse are plotted in gray dots.The missed H I fraction without (f missed ) and with ( f missed,PBa ) VLA primary beam attenuation applied to both single-dish and interferometric data are plotted in red and green triangles, respectively.Galaxy names are given.

4. 4 .
Localized Distribution and Kinematics of the Diffuse H I 4.4.1.Inspecting Moment Images of the Diffuse H I

Figure 6 .
Figure 6.Radial profiles of diffuse H I properties.The profiles are color coded by galaxy name, as denoted in each panel, and the radii are normalized by the optical radius R 25 .All profiles start from the radius equal to the beam FWHM of FAST.All measurements are made at the same resolution of FEASTS (9.1 ± 2.5 kpc).(a): Column density profile.The column densities are not corrected for projection because the inclinations in outer disks where diffuse H I dominates the total H I are highly uncertain.The dashed horizontal line shows the rough column density threshold of selecting the dense H I. (b): Profile of diffuse H I column densities over the total H I column densities.(c): Profile of projection-corrected radial velocity difference between the dense and diffuse H I (former minus latter) measured along the major axis of galaxies (see Section 3.3.1).The projection-corrected radial velocities along the major-axis approximate rotational velocities, but we warn of their uncertainties.The solid and dashed curves are for the receding and approaching side of the disks, respectively.The dotted line marks the position of y = 0. (d): Similar to panel (c), but the radial velocity difference is normalized by the radial velocity of the dense H I along the major axis.Values between zero and unity indicate lagged rotational/orbital velocity (see Section 3.3.1).(e): Profile of velocity dispersion.The solid and dashed lines are for the diffuse and dense H I, respectively.(f): Profile of velocity dispersion over the projection-corrected radial velocity along the major axis for the diffuse H I.

Figure 8 .
Figure 8. Relation between radially distributed H I properties and CGM properties.Different colors are for different galaxies.(a): CGM density as a function of radius, estimated based on scale relations of group CGM properties.(b): Thickness (h H I ) profile derived assuming the diffuse H I is in pressure equilibrium with the CGM.(c): Relation between the diffuse H I thickness and the column density.All dotted lines have a slope of unity.In panels (b) and (c), the estimation based on column densities that are (not) corrected for projection effects are plotted with (solid) dashed curves.

Figure 13 .
Figure13.The moment images of the galaxy NGC 628.The sky region displayed is the full region of the FEASTS data.Row 1: moment 0-2 images of FEASTS data, representing the total H at the FAST resolution.The circle in the bottom left shows the size of the FAST beam.Row 2: moment 0-2 images of THINGS data, representing the dense H I at the VLA resolution.We do not show the VLA synthesis beam as it is too small compared to the field of view.Row 3: moment 0-2 images of dense H I at the FAST resolution.Row 4: moment 0-2 images of the total H I minus the dense H I, representing the diffuse H I at the resolution.Row 5: the difference in H I moment 1 images between the dense H I and diffuse H I at the FAST resolution (the former minus the latter).

Figure 14 .
Figure 14.Same as Figure 13, but for the galaxy NGC 925.

Figure 15 .
Figure 15.Same as Figure 13, but for the galaxy NGC 2841.

Figure 16 .
Figure 16.Same as Figure 13, but for the galaxy NGC 2903.

Figure 17 .
Figure 17.Same as Figure 13, but for the galaxy NGC 3198.

Figure 18 .
Figure 18.Same as Figure 13, but for the galaxy NGC 3521.

Figure 19 .
Figure 19.Same as Figure 13, but for the galaxy NGC 5055.

Figure 20 .
Figure 20.Same as Figure 13, but for the galaxy NGC 5194.

Figure 21 .
Figure 21.Same as Figure 13, but for the galaxy NGC 5457.

Table 2
Observing Information of Targets Studied in This Paper Note.Column (1): galaxy name.Column (2): rms level of the FEASTS cube, in units of mJy per FAST beam.Column (3): H I column density limit of the FEASTS data, assuming 3σ detection and 20 km s −1 line widths.Column (4): rms level of the THINGS cube, in units of mJy per VLA beam.Column (5): channel width of the THINGS data.Column (6): major axis of the THINGS beam.Column (7): minor axis of the THINGS beam.Column (8): H I column density limit of the THINGS data, assuming 3σ detection and 20 km s −1 line widths.Column (9): median signal-to-noise ratio (S/N) of the H I column density image.

Table 4
Cross-calibration Result Notes.Column (1): galaxy name.Column (2): H I flux from the FEASTS data.We do not add error bars because the flux uncertainties are less than 0.01%.Column (3): H I mass from the FEASTS data.Column (4): the fraction of interferometry missed H I flux over the total H I flux ( f missed ) based on W08 cubes.Column (5): the fraction of interferometry missed H I flux over the total H I flux ( f missed ) based on newly reduced THINGS cubes.Column (6): the fraction of diffuse H I over the total H I flux ( f diffuse ) based on newly reduced THINGS cubes.
Catinella et al. 201820)ld reflect noise.Figure2.The change of M H I due to the inclusion of missed H I in the H I mass vs. stellar mass space.The red, green, and blue symbols mark the M H I derived from the FEASTS data, the new THINGS cubes, and W08 THINGS cubes.We link the measurements for the same galaxy with dotted lines.The solid and dashed gray lines show the median and 75 percentile values of M H I , respectively, as a function of M * for general galaxies(Catinella et al. 2018); the solid and dashed cyan lines show the mean relation of M H I vs. M * and its scatter for star-forming galaxies(Janowiecki et al. 2020).(1σ) of the M H I versus M * mean relation of star-forming galaxies(Janowiecki et al. 2020).They are also comparable to half the scatter (50-75 percentiles) of the M H I versus M * relation of general galaxies (xGASS;Catinella et al. 2018).The deviations of M H I from these scaling relations are related to the physical processes that replenish, consume, or remove the H I reservoir in galaxies.We can expect a similar shift of data points around the scaling relation of M H I versus SFR, the