Neutral Hydrogen content of dwarf galaxies in different environments

We select galaxies from the matched sample between the 40% complete Arecibo Legacy Fast Arecibo L-band Feed Array (ALFALFA, a.k.a. $\alpha.40$; Haynes et al. 2011) survey and Seventh Data Releases of SDSS (SDSS DR7; Abazajian et al. 2009).

The SDSS uses a 2.5 m telescope located at Apache Point Observatory (APO) near Sacramento Peak in Southern New Mexico. The main goal is to obtain photometry in five broad bands over 10,000 deg² of high-latitude sky, and spectroscopy of a million galaxies and 100,000 quasars over this same region. SDSS DR7 comprises images and spectra for millions of galaxies, centered at a redshift $\sim$ 0.1. The extinction-corrected spectroscopic data are complete at 17.72 mag in r-band.

In previous studies, it was discovered that the SDSS photometric pipeline tends to overestimate the sky background of galaxies with extended low surface brightness by $\sim$ 0.5 mag (He et al. 2013; Liu et al. 2008; Lisker et al. 2007). This is particularly relevant for dwarf galaxies with low surface brightness as their luminosities are more vulnerable to sky subtraction. To address this issue, we selected a sub-sample of low surface brightness galaxies (LSBGs; Du et al. 2015) in $\alpha.40$, with central surface brightness in the B band $\mu_{0}$(B) $>22.5$ mag arcsec^-2, and recalculated their photometry using the g-band and r-band SDSS mosaic images from the Twelfth Data Releases of the Sloan Digital Sky Survey (SDSS DR12; Alam et al. 2015), implementing improvements in sky-background subtraction following Zheng et al. (2015) and Guo et al. (2020). For galaxies with brighter surface brightness, we utilized Petrosian magnitudes from the released SDSS DR7 catalogue.

The stellar mass is estimated using the stellar mass-to-light ratio formula proposed by Bell et al. (2003)

Here we adopt the Kroupa stellar initial mass function (Kroupa 2002) with $a_{r}=-0.306-0.15$ and $b_{r}=1.097$, $a_{i}=-0.222-0.15$ and $b_{i}=0.864$ (Bell et al. 2003).

The ALFALFA survey is the biggest blind extragalactic H i survey, with a wide sky coverage of 7,000 deg². The spatial resolution beam size is about 3.5 arcmin. The catalogue encompasses 21-cm line spectra of more than 30,000 extragalactic sources with radial velocities $<$18,000 km s^-1. The average integration time of ALFALFA sources is 48s, corresponding to a minimum flux of 0.18 Jy at a velocity width of 50 km/s. Here we use the 40% complete ALFALFA, a.k.a. $\alpha.40$. There are 15,855 H i detected sources in $\alpha.40$, among which 15,041 are extragalactic objects and the rest 814 are possibly to be Galactic high-velocity clouds. Cross-matched with the SDSS DR7, there are 12,423 galaxies with optical counterparts (Haynes et al. 2011).

We select dwarf galaxies following the criteria in Hu et al. (2023). Firstly, we select galaxies with absolute magnitude $\rm M_{r}>-18$ and the H i spectra with high signal-to-noise ratios, S/R${}_{HI}>10$. Then, we visually remove the galaxies which have multiple optical counterparts within $6^{\prime}$ (corresponding to twice the Arecibo beam size), which leaves 770 dwarf galaxies. Secondly, we use the regions where both SDSS and ALFALFA have detection, as shown in Fig. 1. In order to avoid the incompleteness of neighbor galaxies at the edges of the survey, we further restrict to $5^{\circ}<DEC<35^{\circ}$, $125^{\circ}<RA<235^{\circ}$; and $0^{\circ}<DEC<30^{\circ}$, $-25^{\circ}<RA<25^{\circ}$, and we request the samples to be within a redshift range (0.003, 0.0133), leading to 413 selected galaxies. Finally, we restrict our sample to have r-band magnitude brighter than 20 and H i mass $>10^{8.14}M_{\odot}$. This r-band limit is selected to have an optically complete sample set at 95% level ¹¹1From COMBO S11 field, see on SDSS website: https://www.sdss4.org/dr17/imaging/other_info, and this H i mass limit is chosen to ensure an H i mass complete sample in the redshift of interest (Fig. 2). The final sample consists of 320 dwarf galaxies.

The ALFALFA survey has provided a substantial sample of galaxies with detectable H i content. However, due to the relatively short integration time of the survey, the sample mainly consists of H i-rich dwarf galaxies. In order to gain a comprehensive understanding of the variability of H i content in different environments, including H i-poor galaxies, we have increased the integration time and utilized the Five-hundred-meter Aperture Spherical Telescope (FAST; Nan 2006; Jiang et al. 2019) to observe a set of 10 randomly selected dwarf galaxies, which have angular distances to any $\alpha.100$ sources greater than $6^{\prime}$ (twice the Arecibo beam size). These galaxies either have not been detected in the ALFALFA survey or have low signal-to-noise ratios (S/N, as shown in Fig. 2). By expanding our observations to include these targets, we aim to obtain a more comprehensive understanding of the distribution of H i content across various environments.

FAST is a single-dish radio telescope with an effective diameter of 300 meters and the highest sensitivity among ground-based single-dish radio telescopes. FAST is currently mounted with a 19-beam receiver covering a frequency range of 1.05–1.45GHz. Its frequency resolution is about 7.63 kHz (65,536 channels over 1.0–1.5 GHz bandwidth) and the angular resolution is about 2.9 arcmin at low redshift for 21-cm line spectra. The system temperature is $\sim$20K for zenith angles within 26.4 deg observations.

The observations were carried out based on a ‘Shared-Risk’ project, using the position switch ON-OFF mode recording two polarizations (XX and YY) for each target. The ON-source and OFF-source integration time is 120 seconds in most cases, and the overhead time (switch between ON-source and OFF-source positions) is 30 seconds. We have carefully chosen the OFF point for each source so that a side beam (M08 or M14²²2We found that some delta RFI only appeared in the M14 beam so we only used the M14 beam for those M08 positions with contamination) would be pointing to the source when the central beam (M01) is pointing to the OFF point (Zheng et al. 2020). This strategy successfully reduced the overhead time rates down to around 20%. For most of our targets, the sampling time is 0.1 s, we inject the high (10K) noise diode signal for 0.1 s every 1 s during the observation. For AGC 125644, we use a 0.5 s sampling time and inject the high noise diode signal for 2 s every 20 s. The majority of the observations have a total on-source integration time of approximately 19.2 minutes. Details are listed in Table 1.

We perform the following procedures for each beam and their corresponding polarizations (XX and YY).

Flux calibration. We use the high (10K) noise diode other than the standard calibrators for our flux calibrations. First, we convert the recorded noise diode signal into antenna temperature (K). To obtain the flux density in units of Janskys (Jy), we utilized the gain values reported in Table 5 of Jiang et al. (2020), for the M01 and M08 (M14) beams, respectively.

Baseline and standing wave removal. Following Jing et al. (2023)³³3HiFAST: https://hifast.readthedocs.io, we remove a lower-order Polynomial baseline to each OFF-subtracted spectrum (ON-source subtract OFF-source for each beam, i.e. M01 ON-source position minus M01 OFF-source position ⁴⁴4Initially, our plan was to subtract the continuum using the same time from OFF position, specifically the M01 ON-source position minus the M08 ON-source position (pointing the OFF-source position). However, we discovered that the period and phase of the standing wave differed between the different beams, making it challenging to fit the standing wave if we performed the same time OFF subtraction. We noticed that the standing wave varies with time, but we were able to subtract the ripple of the spectra by performing the same beam OFF subtraction.). There are $\sim$ 1 MHz standing waves covering all bandwidth of FAST spectra, which are caused by reflections between the receiver and the bottom panel of the telescope (138.63 meters causing a 1.08 MHz standing wave). We remove the standing wave by fitting a sinusoidal function using spectra around the heliocentric H i velocity of the target.

The observations were conducted in topocentric mode, which means that the reference frame is not a rest frame due to the Earth’s movement. To account for this, we convert the observed frequency to heliocentric velocity. The final spectra are obtained by averaging the spectra from M01 and M08 (M14).

The spectra of detected FAST sources are presented in Fig. 3. Out of these sources, 7 of them have good signal-to-noise ratios (S/N), which allows for reliable analysis of their H i properties. However, for the remaining 3 sources that have low S/N, their spectra only provide upper limits for the analysis. These upper limits are used to estimate the potential H i content in these sources and we incorporate them into the overall analysis.

To illustrate the advantage of using longer integration time with the FAST telescope for H i, we choose two galaxies for comparison from the ALFALFA sample: UGC 8838 and AGC 125644. The spectral profile of UGC 8838 in Figure 4 exhibits a relatively low signal-to-noise ratio (S/N), while AGC 125644 displays a dubious sub-component between 6900 km/s and 7000 km/s, also with a very low S/N. Figure 4 illustrates that, by utilizing a longer integration time, FAST effectively improved the signal-to-noise ratio (S/N) of UGC 8838 from 7.5 to 14.8. For AGC 125644, we confirmed the existence of the suspicious sub-component, despite the S/N in ALFALFA already being as high as 7.1. This discovery led to an expansion of its line width by a factor of two compared to its measurement in ALFALFA. Overall, it demonstrates the improved detection capabilities of the FAST telescope through longer integration times, enabling the identification of previously unidentified sub-components and enhancing the S/N ratio of the observed galaxies.

We future extract the physical properties as follows, with the derived properties listed in Table 2.

Velocity width $\bf W_{\bf 50}$ and $\bf W_{\bf 20}$. The algorithm follows the procedure in Guo et al. (2020). We distinguish between systems exhibiting one peak and two peaks in their H i spectra. For sources with a single peak, we identify the wavelengths at which the fluxes are 50% (20%) of the peak values to obtain the raw values, $W_{50,raw}$ ($W_{20,raw}$). For sources with two peaks, the 50% (20%) fluxes are calculated for each side, and we identify the corresponding characteristic wavelengths at each rising side of the spectrum to measure the velocity width. We perform polynomial fitting at each rising side (Springob et al. 2005; Haynes et al. 2011) to obtain the characteristic wavelengths. We then subtract the instrumental broadening as follows (Catinella et al. 2012):

where $\Delta v$ represents the velocity resolution, which is approximately 1.65 km s^-1.

The heliocentric velocity and H i line integrated flux density, $S_{21}$, in units of Jy km s^-1. The heliocentric velocity is defined as the center velocity of $W_{50}$. The total flux is then computed by integrating the continuum subtracted H i spectrum within the $\pm 2\times W_{50}$ range from the heliocentric velocity. The 5 $\sigma$ upper limits of the flux density for three undetected galaxies (IC 3487, IC 3663, and UGC 7854) are calculated as $5\sqrt{W_{50}/1.65\times\sigma_{rms}^{2}}\times 1.65$ under the assumption that $W_{50}=100$ km/s.

Signal to noise ratio (S/N) of the spectrum. We perform two different estimations of S/N:

(1) An S/N estimation analog to ALFALFA following Haynes et al. (2011), the S/N of the detection is estimated as:

where $\omega_{smo}$ is a smoothing width set to be $W_{50}/20$ for $W_{50}<$ 400 km s^-1, and to be 20 for $W_{50}\geq$ 400 km s^-1, and $\sigma_{rms}$ is the root-mean-square of the baseline.

(2) An average S/N within velocity width $W_{50}$:

H i mass and distance. In ALFALFA, the distances for galaxies with $\rm cz>6000$ km s^-1 are the Hubble distances, estimated as $\rm d=cz/H_{0}$ where $H_{0}$=70 km s^-1 Mpc^-1 is the Hubble constant, while for galaxies with $cz<6000$ km s^-1, the distances are obtained with a combination of primary distances from the literature and secondary from the TFR. For our newly observed FAST targets, we only use the primary distances in the literature, as they are more accurate. When no accurate distance measurements are available in the literature, we use Hubble distance instead which is calculated from the heliocentric velocity of the H i spectrum. The H i mass of the galaxy is then derived from the integrated H i flux and distance following Roberts (1962), under the assumption that the H i content is optically thin:

where d is the luminosity distance to the galaxy unit in Mpc.

We make use of the Illustris-TNG cosmological hydrodynamical simulations⁵⁵5https://www.tng-project.org for the comparison. TNG50 (Nelson et al. 2019; Pillepich et al. 2019) traces $2\times 2160^{3}$ dark matter particles and gas cells in a period box size of 51.7 Mpc. The mass resolution is $4.5\times 10^{5}M_{\odot}$ and $8.5\times 10^{4}M_{\odot}$ for dark matter and baryon, respectively. Stellar masses are calculated using all star particles within each subhalo. We adopt the “Neutral Hydrogen Abundance” from the catalogue for H i mass in non-star-forming cells and calculate the cold hydrogen gas mass (H i $+\rm H_{2}$) using the two-phase interstellar medium model (a modification of Springel & Hernquist 2003) in star-forming cells. Then, we drop out the molecular hydrogen from all cold hydrogen mass to obtain the H i mass for star-forming cells. The molecular hydrogen fraction $f_{H_{2}}$ in star-forming regions is calculated by the modified KMT model (Krumholz et al. 2009; McKee & Krumholz 2010) as

where s is given by

and

where $Z$ represents the metallicity for gas cell in units of solar metallicity $Z_{\odot}=0.0127$, $\sigma=Z\times 10^{-25}\ \rm{m^{2}}$ is the dust cross-section and $\mu=3.9\times 10^{-27}\ \rm{kg}$ is the mean mass per particles. The Jeans length approximation method is adopted when estimating the gas surface density $\Sigma_{gas}$. We discard galaxies with less than 200 star particles to ensure a reliable estimation of stellar mass.

In order to compare to observations, here we select dwarf galaxies with $\rm M_{r}>-18$ and H i mass $\rm>10^{8.14}M_{\odot}$ for the main comparison (TNG50 gas-rich dwarf galaxies). While we did not set a lower limit for the r-band luminosity, Figure 5 demonstrates that the distributions in $\rm M_{r}$ are similar between the $\alpha.40$ dwarf sample and the TNG50 gas-rich dwarf galaxy sample. Furthermore, to examine the impact of environmental effects, we utilized a larger sample by including all galaxies with $\rm M_{r}>-18$, which encompasses both gas-rich and gas-poor galaxies. As depicted in Figure 5, it is evident that such selection also encompasses a broader range of faint galaxies in terms of r-band magnitude.

We adopt two kinds of environment definitions, the projected number density and the distance to the nearest group/cluster.

The projected number density is defined as the number of galaxies brighter than -16.6 mag (r-band) within 1 Mpc projected distance and up to $\pm 1000$ km/s velocity difference along the line-of-sight. Here we use the spectroscopic data of SDSS which is complete at 17.72 mag in r-band, corresponding to an absolute magnitude of $\sim-16.6$ mag at the redshift of 0.017 (z=0.0133$+\frac{1000km/s}{c}$).

Distance to the nearest group is defined as the minimum value of the distance to the group center in a unit of the corresponding group viral radius (Guo et al. 2020). The group catalogue is taken from (Saulder et al. 2016) which applies a friends-of-friends (FoF) group finder algorithm on the SDSS DR12 (Alam et al. 2015) and the 2MASS Redshift Survey (2MRS; Huchra et al. 2012). Compared to the group catalogue by Yang et al. (2007), the group catalogue used in this paper includes more groups in the local Universe (including groups at $z<0.01$). Various observational biases have been taken into account, including the Malmquist bias, the ‘Fingers of God’, etc. The groups/clusters masses are derived from the total luminosity, luminosity distance, velocity dispersion, group radius, and the number of detected group members (we refer the reader to Saulder et al. 2016 for more details). This method has been precisely calibrated on the mock catalogues from the Millennium simulation (Guo et al. 2011, 2013). The minimum and median distances to the nearest group of the parent dwarf galaxies are 0.074 and 4.339, respectively.

The histograms of two different definitions of the environment are shown in Fig. 6. Most $\alpha.40$ dwarf galaxies reside in relatively low-density regions, outside the virial radius of surrounding clusters/groups. Galaxies at high densities or close to groups/clusters suffer from environmental stripping and likely lose their H i gas in a short time scale. One needs a longer integration time to reach deep detection of galaxies in such environments. FAST, with its longer integration time, captures a higher fraction of galaxies in these high-density regions, as depicted by the red histogram. Accordingly, the combination of the gas-rich $\alpha.40$ galaxies and the dwarf galaxies residing in denser regions, exposed for a longer duration through FAST, form a valuable sample for investigating environmental dependencies.

Figure 7 presents the relationship between H i mass and stellar mass. The H i mass of the jointly selected gas-rich dwarf galaxy sample appears to increase with their stellar masses. The observed H i v.s. stellar mass relation is well reproduced in simulations when taking into account the detection limit of the $\alpha.40$. However, the gas-poor galaxies detected by FAST push this H i mass v.s. stellar mass relation downwards more than an order.

When including both gas-rich and gas-poor dwarf galaxies in simulations, it shows an even stronger dependence of the H i mass with stellar masses. Dwarf galaxies with low H i masses as detected by FAST are also found in simulations. Indeed, TNG50 exhibits a continuous distribution along the y-axis, with a relatively small fraction of galaxies having $\rm logM_{HI}<7.5$ at $\rm logM_{*}\sim 8.5$.

Interestingly, we observe a common boundary for the maximum $\rm M_{HI}$ (H i mass) in both our observational data and simulations. This boundary sets a limit on the highest H i mass that can exist for a given stellar mass ($\rm M_{*}$). Such behavior also translates into the $\rm M_{HI}$ to $\rm M_{*}$ ratio as a function of stellar mass (right panel). To describe this upper boundary, we employed a fitting function as follows.

This is equivalent to

Given the detection limit, at least 97.2% $\alpha.40$ dwarf galaxies are below this boundary, while in simulations for TNG50 dwarf galaxies, 99.2% are below this boundary.

Figure 8 displays the distribution of the H i-to-stellar mass ratio ($\rm M_{\rm HI}/M_{*}$) across different environments. To extend the sample size, we include data from two Local Volume dwarf galaxy samples: Local Irregulars That Trace Luminosity Extremes, The H i Nearby Galaxy Survey (LITTLE THINGS; Oh et al. 2015) and Survey of H i in Extremely Low-mass Dwarfs (SHIELD; McNichols et al. 2016).

The left panels of the figure demonstrate a slight decrease in the $\rm M_{\rm HI}/M_{*}$ as the distance to the nearest group decreases. The black lines in the left panels indicate where the distance to the nearest group is three times the group virial radius, beyond which the high-density environmental effects typically vanish. In the right panels, the black lines correspond to the median values of the number density for the $\alpha.40$ dwarf galaxies (3) and TNG50 dwarf (1). A consistent trend is observed, indicating a decrease in the $\rm M_{\rm HI}/M_{*}$ as the environment becomes denser, regardless of the specific definition of the environment. To remove the mass dependence, Fig. 9 shows the $\rm M_{\rm HI}/M_{*}$ vs. $\rm M_{*}$ relations for galaxies in the 25% most dense and the 25% most isolated environments. It is clear that galaxies in dense regions have a lower $\rm M_{\rm HI}/M_{*}$ at any given stellar mass.

We observe a lower boundary of the $\rm M_{\rm HI}/M_{*}$ ratio at around 0.1, which could be an artifact resulting from the limited observation depth of the $\alpha.40$ survey. At low densities galaxies detected by FAST reside in the same region as the $\alpha.40$ sample. However, the high sensitivity of FAST, along with longer integration times, extends the lower boundary by a factor of 10 near groups/clusters. These findings suggest that there is a gradual and continuous process of H i gas stripping around groups/clusters, rather than a sudden removal.

Such environmental dependence is much weaker, if any, in simulations. In the case of TNG50 gas-rich dwarf galaxies which mimic the observed $\alpha.40$ dwarf sample, their $\rm M_{\rm HI}/M_{*}$ ratio does not vary much when galaxies get closer to groups/clusters, or when the density increases. This suggests that gas stripping occurs faster in simulations than those in the real Universe. When including gas-poor dwarf galaxies in TNG50, it encloses the regions covered by the gas-poor dwarf galaxies detected through the deep FAST observation. The dependence on environments is stronger compared to the TNG50 gas-rich sample, but overall, the dependence remains weak.

In summary, the process of gas stripping from galaxies as they approach high-density regions is rapid, but the timescale of this process could be even shorter in simulations than those in observations.

Previous studies found that H i-to-stellar mass ratio is closely related to galaxies’ optical properties (Zhang et al. 2009; Catinella et al. 2010; Li et al. 2012; Zhang et al. 2021b). In Figure 10, we present the relation between $\rm M_{\rm HI}/M_{*}$ and optical proprieties, as proposed by Zhang et al. (2009). In the upper left panel, we show the $\rm M_{\rm HI}/M_{*}$ as a function of the combination of the g-r color and stellar surface mass density. The stellar surface mass density is defined as $\rm log(\mu_{*,i})=log(M_{*})-log(2\pi R^{2}_{50,i})$, where $\rm M_{*}$ represents the total stellar mass, and $\rm R_{50,i}$ corresponds to the i-band Petrosian half-light radius. In the upper right panel, the stellar surface mass density is replaced with surface brightness, which is defined as $\rm\mu_{i}=m_{i}+2.5log(2\pi R^{2}_{50,i})$, where $m_{i}$ represents the apparent Petrosian i-band magnitude. Zhang et al. (2009) used a sample of 800 galaxies from the SDSS DR4 and the HyperLeda H i (Paturel et al. 2003) catalogue (Sample II) and obtain the scaling relations as presented by the solid lines, with the dashed lines corresponding to the 1$\sigma$ scatter.

Figure 10 demonstrates that the H i-selected sample from $\alpha.40$ exhibits a slightly higher $\rm M_{\rm HI}/M_{*}$ compared to the sample studied by Zhang et al. (2009). This divergence could be attributed to the exclusion of gas-poor galaxies in $\alpha.40$, which are primarily found in high-density regions. To address this, we incorporate the inclusion of high-density gas-poor galaxies detected by the FAST telescope, particularly in high-density environments. When analyzing the samples separately based on different environments, a clear dependence on the environment becomes apparent. For a given combination of the optical properties, galaxies in dense regions seems to have lower $\rm M_{\rm HI}/M_{*}$.

In the lower panels of Fig. 10, we present a comparison with the TNG50 gas-rich dwarf (left) and TNG50 dwarf galaxies (right). For gas-rich dwarf galaxies, it shows a dependence on environments mainly on the right-hand side where the gas fraction is relatively high. When including gas-poor dwarf galaxies, $\rm M_{\rm HI}/M_{*}$ reaches lower values at the left-hand side, galaxies in dense regions are presented well below the scaling relations found by Zhang et al. (2009).

These findings imply that the current understanding of the relationship between the $\rm M_{\rm HI}/M_{*}$ and optical properties may not be accurate enough without considering the influence of environmental factors.