THE SEARCH FOR Hi EMISSION AT z ≈ 0.4 IN GRAVITATIONALLY LENSED GALAXIES WITH THE GREEN BANK TELESCOPE

L. R. Hunt; D. J. Pisano; S. Edel

doi:10.3847/0004-6256/152/2/30

1. INTRODUCTION

In order to better understand the star formation history of the universe we must study the three indicators of star formation and their sustainability as a function redshift: changes of stellar mass, star formation rate, and gas content of galaxies. Studies of stellar mass density between redshifts, z, of zero and four have shown that approximately 10% of today's stellar mass has formed at z ≈ 3, and 50% to 75% has formed at z ≈ 1 (Dickinson et al. 2003; Fontana et al. 2003; Rudnick et al. 2003) with the total stellar mass density increasing by an order of magnitude between z = 3.5 and z = 0.1 (Pérez-González et al. 2008; Marchesini et al. 2009; Ilbert et al. 2013). Like stellar mass density, the star formation rate has been shown to increase with increasing redshift, up to an order of magnitude between z = 0.01 and z = 1, peaking between z = 2 and 3 (Flores et al. 1999; Haarsma et al. 2000; Wilson et al. 2002; Glazebrook et al. 2004; Hopkins 2004; Hopkins & Beacom 2006). Although stellar mass density and star formation rate density have similar trends, the predicted stellar mass density from instantaneous star formation rate density measurements is higher than the observed stellar mass density by approximately 60% (Madau & Dickinson 2014). Studies of molecular gas, the material from which stars form, have been biased toward gas-rich, actively star-forming galaxies, but indicate that the gas mass fraction increases over redshift and peaks at z ∼ 2, similar to the star formation rate (Carilli & Walter 2013). Neutral atomic hydrogen (Hi) in galaxies is the ultimate fuel for star formation, but the Hi content of galaxies has only been measured in damped Lyα systems beyond z = 2 (Noterdaeme et al. 2012), and indirectly between z ∼ 0.25 and z = 2. Saintonge et al. (2011) found little correlation between M_Hi from Catinella et al. (2010) and ${M}_{{{\rm{H}}}_{2}}$ in massive galaxies in the CO Legacy Database for the GALEX Arecibo SDSS Survey. Michałowski et al. (2015) found that galaxies hosting long gamma-ray bursts are deficient in molecular gas but abundant in Hi, suggesting that at least the initial burst of star formation could come directly from the atomic gas. So although we may have a glimpse at how molecular gas changes as a function of redshift, we are still unsure how the atomic gas, still an important element of star formation, changes as a function of redshift between z = 0 and z = 1.

To date, atomic gas between z = 0 and z = 0.2 has been studied by measuring the 21 cm Hi emission. Surveys such as Arecibo Legacy Fast ALFA (ALFALFA) survey and the HI Parkes All Sky Survey (HIPASS) have detected large samples of galaxies and measured their Hi content out to z ∼ 0.08 (Zwaan et al. 2003; Giovanelli et al. 2005), but until recently little was known about Hi 21 cm emission beyond z = 0.1. Zwaan et al. (2001) used the Westerbork Synthesis Radio Telescope (WSRT) to make the first detection of Hi at z > 0.1, z = 0.1766, finding M_Hi = (6.0 ± 0.8) × 10⁹ for a galaxy in the cluster Abell 2218. Catinella & Cortese (2015) and Verheijen et al. (2010) detected Hi in ∼180 galaxies between z = 0.16–0.25 down to masses of M_Hi = 3 × 10¹⁰ to 2 × 10⁹ M_⊙ with the Arecibo telescope and the WSRT, respectively.

Indirect detections of Hi have been made to z = 0.8 using observing techniques such as stacking (Lah et al. 2009) and intensity mapping (Chang et al. 2010). Lah et al. (2009) found average Hi mass of (6.6 ± 3.5) × 10⁹ M_⊙ per galaxy in Abell 370, a cluster at z = 0.37. Numerous groups have used the 100 m Green Bank Telescope³ (GBT), the only telescope with a cooled receiver that can detect Hi at redshift z ≥ 0.45 with a reasonable integration time, to create an Hi intensity map (Chang et al. 2010; Masui et al. 2013; Switzer et al. 2013). After cross-correlating the GBT data with optical data from the WiggleZ Dark Energy Survey, Masui et al. (2013) made a 7.4σ detection of Hi density, Ω_Hi = (0.4 ± 0.05(stat.) ± 0.04(sys.)) × 10⁻³ × (1/rb), where r, the stochasticity, and b, the bias, are not well constrained.

Direct detections beyond z = 0.25 are difficult due to the weakness of the 21 cm H i line, the limited sensitivity and frequency coverage of present-day radio telescopes, and the many sources of radio frequency interference (RFI) in the frequency bands that cover redshifted Hi emission. After successful pilot observations (Fernández et al. 2013), the COSMOS Hi Large Extragalactic Survey is currently using the recently upgraded Karl G. Jansky Very Large Array to search for 21 cm emission in individual galaxies in the well observed COSMOS field out to z = 0.45. Telescopes designed specifically to carry out intensity mapping surveys, like the Canadian Hydrogen Intensity Mapping Experiment (Battye et al. 2013), Baryon Acoustic Oscillations in Neutral Gas Observations (Bandura et al. 2014), and Tianlai (Chen 2015) will greatly improve measurements of Ω_Hi. Planned deep field surveys with telescopes such as ASKAP, MeerKAT, and the Square Kilometer Array (Meyer 2009; Holwerda et al. 2012; Staveley-Smith & Oosterloo 2015) will have the sensitivity and frequency coverage required to directly detect Hi emission in individual galaxies beyond z = 0.45.

Until these telescopes are completed, direct detections of magnified Hi 21 cm emission from strongly lensed sources can be made to z ∼ 1 using current telescopes (Deane et al. 2015). This technique has been used to detect magnified CO emission in many strongly lensed galaxies, the first being a galaxy at z = 2.2867 by Brown & Vanden Bout (1991), and more recently a survey detecting emission from sources beyond z = 4 (Vieira et al. 2013). Probing lensed sources for Hi emission should also be possible, but careful concern is required when selecting these sources to ensure they are sufficiently magnified and the observed frequencies are not saturated with RFI. In this study we report on our observations of three gravitationally lensed galaxies behind the galaxy cluster Abell 773 (Sand et al. 2005), chosen because they had known redshifts and were likely to be highly magnified.

Throughout this paper we assume H_o = 69.7 km s⁻¹ Mpc⁻¹, Ω_M = 0.282, and Ω_Λ = 0.718 (Hinshaw et al. 2013) and use the cosmology calculator from Wright (2006)⁴ to calculate distances. In Section 2 we explain our observations, data reduction method, and the flagging method we used to remove RFI. In Section 3 we present the results. In Section 4 we discuss our results, the effectiveness of our flagging, and mass estimates based on galaxy magnitudes. In Section 5 we go through our conclusion and briefly discuss the application of our technique to additional targets.

2. OBSERVATIONS AND ANALYSIS

2.1. Sources

Three sources were selected, F3, F13, and F18, from a list gravitationally lensed galaxies behind various massive clusters observed with the Hubble Space Telescope (HST; Sand et al. 2005). These observations were taken using the Wide Field Planetary Camera 2 on the HST with the F702W filter. Sand et al. (2005) lists the magnitude of F3 as 21.21 ± 0.02, F13 as 21.52 ± 0.03, and F18 as 23.39 ± 0.11. F3 and F13 both lie at z = 0.398 and F18 lies at z = 0.487 (Sand et al. 2005; Richard et al. 2010). The sources are labeled in Figure 1, which shows the they all fall within the ∼13' GBT beam.

**Figure 1.** An *HST* F814W image of the galaxy cluster Abell 773. The lensed galaxies F3, F13, and F18 are labeled and outlined. The positions for each source can be found in Sand et al. (2005). The lensed galaxies should clearly fall within the 13' GBT beam.
Download figure:
Standard image High-resolution image

We originally targeted these sources because they fell within 0.25 ≤ z ≤ 1.0 and had a high length-to-width ratio which suggested a higher magnification of ∼10. After our observations we obtained a lens model from J. Richard (2009, private communication) which yielded magnifications of 1.7 ± 0.1 for F3, 2.0 ± 0.2 for F13, and 2.7 ± 0.3 for F18.

2.2. Observations

Observations of the three galaxies were carried out at night in order to minimize RFI, over seven years using the GBT which has a gain of 2 K Jy⁻¹. Observations were made on 2007 January 21 and 26, over eight sessions from 2008 January 16–24, and over four sessions from 2014 February 2–6. We used the Prime Focus 2 (PF2) receiver (901–1230 MHz), and the Spectral Processor backend with two polarizations and over two frequency bands for all observations. The first band was centered at 955 MHz (z = 0.487) and the second at 1016 MHz (z = 0.398), each with a 10 MHz bandwidth (Δv = 3140 km s⁻¹ at 955 MHz and Δv = 2952 km s⁻¹ at 1016 MHz). In 2007 and 2014 our data had 512 channels per frequency band for a frequency resolution of 19.5 KHz, and a velocity resolution of 6.2 km s⁻¹ at z = 0.487 and 5.8 km s⁻¹ at z = 0.398. In 2008 the configuration changed to 256 channels per frequency band with the final velocity resolution listed in Table 1. The beam size for the PF2 receiver is ∼13 farcm 6 at 955 MHz and ∼12 farcm 8 at 1018 MHz; the three sources we observed were within 1' of our pointing direction so that all of the emission falls within the beam.

Table 1. Summary of Observations

	z = 0.398 (ν = 1016 MHz)	z = 0.487 (ν = 955.5 MHz)
Pointing Coordinates (J2000)	09^h17^m5462, 51°43'446
Observing Time (hr)	77.6
Effective Integration Time (hr)	17.2	18.1
Bandwidth (MHz)	10
Frequency Resolution (kHz)	39.1
Velocity Resolution (km s⁻¹)	11.5	12.3
System Temperature (K)	26.2	23.7
T_cal (K)	1.61	1.58
% Flagged	6	0.85

Note. Observing time is the total telescope time. Effective integration time is per polarization. the frequency and velocity resolutions listed are the final values after smoothing. T_cal is the temperature injected by the noise diode used to compare what is measured by the telescope to a known value for calibration. The percentage of data flagged includes integrations that were dropped because of high noise, and the percentage of data flagged in Fourier space.

Download table as: ASCII Typeset image

The GBT is located in the National Radio Quiet Zone which greatly reduces most, but not all, RFI from terrestrial sources. These sources of RFI cause both wide- and narrowband interference with a wide range of flux density and will be further discussed in Section 2.4. The Spectral Processor is ideally designed to handle the wide range of flux densities with a dynamic range of 45 dB.

Standard position switching was used for the observations, with a short integration time of two seconds to reduce the number of spectra affected by wideband RFI. In 2007 we spent two minutes on source and two minutes off source, and switched to five minutes on source and five minutes off source for the 2008 and 2014 observations to reduce overhead. The off source pointing was carried out for 2 minute of R.A., or 30 arcmin, ahead of the source for the first year and for 5 minute of R.A. ahead of the source for the rest of the observations so that the same azimuth and zenith angle were tracked as the on source observations. Our total observing time was 77.6 hr, giving an integration time of 17.2 hr at z = 0.398 and 18.1 hr at z = 0.487.

2.3. Data Reduction

Reduction of the data was performed using GBTIDL.⁵ For each night, spectral window, on–off scan pair, polarization, and integration we used the getps procedure which calibrates a total power, position switched scan pair as defined in Equation (1) where T_sys is listed in Table 1, the gain for the GBT is 2 Jy K⁻¹ and the atmospheric opacity correction is τ = 0.01:

$\begin{eqnarray}&&S=\displaystyle \frac{\mathrm{On}-\mathrm{Off}}{\mathrm{Off}}\times {T}_{\mathrm{sys}}\times \;\mathrm{Gain}\times \;{e}^{\tau }.\end{eqnarray} \tag{ 1 }$

Next, a fifth order polynomial baseline, the lowest order polynomial that produces flat baselines, was fit across the frequencies 950.24–953.65 MHz, 954.15–957.25 MHz, and 957.95–960 MHz at z = 0.487, and 1011.25–1011.75 MHz and 1012.5–1021 MHz at z = 0.398. This fit was applied to every integration whether wideband RFI was present or not. The fifth order polynomial applied across our 10 MHz bandwidth removes variations in the bandpass of roughly 2 MHz and should not affect a typical galaxy with a line width of 200 km s⁻¹ (Papastergis et al. 2011) which would have a frequency width of ≲677 kHz. We developed an automated flagging routine, described in more detail in Section 2.4, which we then used to flag all RFI. A spectrum for each night and spectral window was created by accumulating integrations from all on–off scan pairs and both polarizations. The effective integration time at 1016 MHz is much lower, and the percentage of data flagged much higher, because the data for one polarization on night two had abnormally high noise, and were flagged completely. There is no obvious reason why the noise was higher in that one polarization. Next, we used a Gaussian curve to smooth over the first two nights and the last four nights to change the frequency resolution from 19.5 kHz per channel (a velocity resolution of 6.2 km s⁻¹ at z = 0.487 and 5.8 km s⁻¹ at z = 0.398) to 39.1 kHz per channel (a velocity resolution of 12.3 km s⁻¹ at z = 0.487 and 11.5 km s⁻¹ at z = 0.398) to improve the noise and match the other eight nights. A fifth order polynomial baseline was fit to the final spectrum for each night and then we accumulated and averaged all nights to obtain the final spectrum.

2.4. RFI and Flagging Routine

The observed frequency bands contain a large amount of RFI which we believe is caused by airplanes' distance measuring equipment (DME) radar. The ground-to-air portion of the radar transmits between 962 and 1024 MHz. Fisher et al. (2005) describes DME as a pair of strong pulses between an airplane and a ground station sent at 24–30 pairs a second, meaning that each pulse is much shorter than the two second integration time. Our two second integration time is a limit of the Spectral Processor, but future observations will be able to take advantage of the increased dynamic range and processing power of the Versatile GBT Astronomical Spectrometer, allowing for increased bandwidth and spectral resolution, and shorter integration time. The wideband interference could come from a strong, intermittent DME signal outside the observing band, appearing consecutively in up to 77 integrations. The time the RFI was visible in the data ranged from 2–144 s, but was most frequently visible for ∼4 s at a time. The wideband RFI has a characteristic width of 1 MHz which is still larger than the aforementioned 677 kHz frequency expected for the signal.

Both wideband and narrowband interference were removed using a custom sigma-clipping method (Yahil & Vidal 1977), measuring the standard deviation of the spectrum and removing points above or below some multiple of that value. Channels around the clipped point were then blanked to remove the whole spike. This was performed on the frequency spectrum (the frequency domain) to remove narrowband interference, and on its Fourier transform (the Fourier domain) to remove wideband interference.

We only wanted to flag spectra in the Fourier domain when wideband RFI was present in order to avoid unnecessarily removing data. The data containing wideband interference often had tall spikes in the Fourier transform, and we used this to search for wideband interference. We also found that the data that contained narrowband interference had tall spikes in Fourier space. From data analysis we saw that the narrowband interference occurred most frequently at 954 and 958 MHz at z = 0.487, and 1012 MHz and 1018 MHz at z = 0.398. Before we carried out our preliminary Fourier transform to search for wideband RFI, we quickly interpolated over those frequencies. In this way we were able to test for wideband interference using the Fourier domain while avoiding confusion with narrowband interference. If a peak in Fourier space was measured over our threshold, a flux density greater than 0.015 Jy or less than −0.015 Jy, we assume that the integration contained wideband interference.

We then started again with the original spectrum and continued our clipping routine. The values were selected after measuring the maximum and minimum values in the Fourier domain of many spectra both containing and lacking wideband RFI. The unaltered spectra were Fourier transformed and clipped, setting channels larger than 4.3σ to zero and doing the same for four channels on either side when the band had 512 channels and two channels on either side when the band had 256 channels to ensure the spike was removed. The value of 4.3σ and the number of channels flagged on either side of the spike were both chosen after testing various combinations to determine the lowest combination that removed all wideband interference. After the spikes in the Fourier domain were removed, the spectrum was inverse Fourier transformed, and the wideband interference was no longer present in the frequency domain. We continued in the frequency domain, measuring the standard deviation across the central 2 MHz, blanking channels larger than 3.5σ, and removing seven channels on either side when the band had 512 channels, and four channels on either side when the band had 256 channels, eliminating most of the narrowband interference in the frequency domain. Because the narrowband RFI is ever present at 954 and 957.5 MHz, we could not remove all of it from the final spectrum. Again the value for sigma-clipping and width were determined by testing multiple scans to find the lowest combination of values to remove as much of the spike as possible without removing signal.

A test was developed to check the effectiveness of our flagging routine. We used a GBT observation of the galaxy NGC 5375 in which the signal from the galaxy is not visible in a single integration, but becomes visible when many integrations are averaged together. NGC 5375 is at much lower redshift and has a velocity width of 280 km s⁻¹, so its frequency width, 1.4 MHz, is 2.1 times larger than the assumed frequency width of the high redshift sources, 677 kHz. To make sure the Hi signal is not visible in a single integration, we artificially increased the noise in each channel in each integration by adding a random number to the measured flux density in each channel. The signal-to-noise ratio in each integration becomes 0.66 after adding this artificial noise, making the source undetectable in a single integration. Then we introduced wideband interference, extracted from our original data set by fitting a high order polynomial, to approximately 10% of the integrations in the test data set at random. It is important to note that we only flagged about 190 channels, much less than 1% of the data. Next we flagged the data in the Fourier domain for the integrations in which we introduced wideband interference, zeroing any values higher than 4.3 times the noise, as we did to the sources behind Abell 773, and did nothing to the others. We averaged all of the integrations to create a final average spectrum, and compared them to the unaltered final spectrum. The results in Figure 2 show that the Hi signal looks the same when there is no wideband interference, and when the artificial wideband interference is removed using our flagging routine. The spectrum without wideband interference added had an rms of 0.0093 Jy with an integrated signal-to-noise ratio of 8.2, and the spectrum with wideband interference added and then removed had an rms of 0.0095 Jy with an integrated signal-to-noise ratio of 7.6. The rms of the residual spectrum is 0.002 Jy. The signal-to-noise ratio and frequency width of the test source are much larger than those expected from our data, so the test represents an extreme and the flagging routine should have a smaller effect on our data.

**Figure 2.** The spectrum we used to test our flagging routine. The top plot shows the final spectrum without adding wideband interference. The middle plot shows the final spectrum after adding wideband interference to 10% of the integrations. The third plot shows the final spectrum after adding wideband interference and then applying our flagging routine. The final plot shows the difference between the spectrum in which no interference was added, and the spectrum which was flagged in Fourier space.
Download figure:
Standard image High-resolution image

3. RESULTS

The final spectra are shown in Figure 3. The spectra have an rms of 211 μJy at z = 0.487 and 204 μJy at z = 0.398. For comparison, the theoretical values for the noise are 183 μJy and 171 μJy, respectively. This is calculated from

$\begin{eqnarray}&&\sigma =\displaystyle \frac{{T}_{\mathrm{sys}}}{G\;\sqrt{{N}_{\mathrm{pol}}\;{\rm{\Delta }}\nu \;{t}_{\mathrm{eff}}}}\end{eqnarray} \tag{ 2 }$

where T_sys is the system temperature, 23.7 K at ν = 1016 MHZ and 26.2 K at ν = 955 MHZ, G = 2 K Jy⁻¹ is the gain, N_pol = 2 is the number of polarizations, Δν = 39.1 KHz is the frequency width per channel, and t_eff is the effective integration time listed in Table 1. The corresponding M_Hi detection limit is

$\begin{eqnarray}&&{\sigma }_{{M}_{{\rm{H}}{\rm{I}}}}=\displaystyle \frac{2.36\times {10}^{5}{D}_{L}^{2}{\sigma }_{s}{dv}\sqrt{N}{\sigma }_{{\rm{S}}/{\rm{N}}}}{(1+z)\mu }\end{eqnarray} \tag{ 3 }$

where ${\sigma }_{{M}_{{\rm{H}}{\rm{I}}}}$ is the upper limit of the neutral hydrogen mass, D_L is the luminosity distance to the object, 2.18 Gpc at z = 0.398 and 2.77 Gpc at z = 0.487, σ_s is the rms per channel in the spectrum in units of Jy, dv is the channel width in km s⁻¹, N is the number of channels the galaxy would span, σ_S/N is the signal-to-noise ratio, z is redshift, and μ is the magnification. To set a mass limit we need to select a value for N. We use the mode of the line width, ∼200 km s⁻¹, from 10744 galaxies in the ALFALFA survey (Papastergis et al. 2011) divided by a velocity resolution of 12.3 km s⁻¹ at z = 0.487 and 11.5 km s⁻¹ at z = 0.398 to find N = 16 and 17, respectively. Using the above parameters, we calculate the 3σ detection limit to be M_Hi = 1.50 × 10¹⁰ M_⊙ and M_Hi = 6.36 × 10⁹ M_⊙ respectively.

4. DISCUSSION

4.1. Effectiveness of Flagging

The flagging procedure we used was possible only because the galaxy is not visible in any single integration and there was no danger of clipping or altering our signal. As mentioned in Section 2.2, we measured the standard deviation of the central 2 MHz (628 km s⁻¹ at 955 MHz and 590 km s⁻¹ at 1016 MHz) of the spectrum in both bands and used that value for sigma-clipping. We did this because that region of the spectrum was generally devoid of narrowband RFI. In the central 2 MHz very few integrations were clipped in the frequency domain, so apart from the one polarization on night two that was discarded due to high noise, most of the information that was lost came from clipping in the Fourier domain. Only 6.53% of the data were flagged at 1016 MHz and 1.6% were flagged at 955 MHz. If we ignored integrations with wideband RFI instead of flagging in Fourier space, we would have removed ∼14% of the data at 955 MHz and ∼10% of the data at 1016 MHz. While this is a small effect on the final noise of the spectrum, it also yields cleaner baselines.

**Figure 3.** The final combined spectra for our data at both z = 0.398 and z = 0.487. The RFI present at 954 MHz and 957.5 MHz persisted no matter how we changed our flagging routine.
Download figure:
Standard image High-resolution image

In Figure 4 we show three examples of spectra before and after flagging. The top panel shows a spectrum with wideband RFI before and after flagging. In the first spectrum, our routine was effective at removing the wideband RFI and we were also able to remove narrowband spikes. The middle panel has a spectrum showing only narrowband RFI. The spectrum was not changed in Fourier space and the narrowband RFI was removed with sigma-clipping. The bottom example in Figure 4 shows flagging performed when the spectrum does not appear to contain any RFI. The spectrum before and after the flagging routine remained exactly the same.

Figure 5 shows the channels that were clipped most often in the frequency domain. In the band centered at 955 MHz (z = 0.398) narrowband interference was frequently present near 954 and 958 MHz. After changing our widening parameters we were still unable to remove it entirely from our final spectrum and it is still visible in Figure 3. We removed many channels on the edge and around 1012 MHz in the band centered at 1016 MHz (z = 0.487). A large percentage of the data are flagged because of narrowband RFI outside the central region. The dips in the middle of each plot in Figure 5 correspond to the areas where we measured the standard deviation and retained most of the data.

4.2. Mass Limits

We were able to use the scaling relation between R band absolute magnitude and Hi mass described by Dénes et al. (2014) derived from HIPASS, to estimate the Hi mass for each galaxy. A galaxy type for these sources was not available so we assumed all three were Sbc spirals, the best case scenario for detection. The estimated masses are ${M}_{{\rm{H}}{{\rm{I}}}_{{\rm{F}}3}}=(2.2\pm 1.3)\times {10}^{9}$ , ${M}_{{\rm{H}}{{\rm{I}}}_{{\rm{F}}13}}=(1.5\pm .8)\times {10}^{9}$ , and ${M}_{{\rm{H}}{{\rm{I}}}_{{\rm{F}}18}}=(4.7\pm 2.3)\times {10}^{8}$ . The calculated average mass weighted by the magnification for F3 and F13 is ${M}_{{\rm{H}}{{\rm{I}}}_{{\rm{F}}3+{\rm{F}}13}}=(1.8\pm 0.7)\times {10}^{9}$ . We obtained the local unmagnified R band magnitude by transforming the magnified, redshifted F702W magnitude using the k-corrections and color relations from Fukugita et al. (1995). F3 and F13 lie at the same redshift so we add the mass of the sources together weighted by their magnification for a total estimated mass detectable at z = 0.398. These mass estimates are ∼30 and 3.7 times smaller than the detection limit for F18 and both F3 and F13, respectively. To bring the noise down to the level required to detect emission from F3 and F13, we require approximately 200 extra hours of integration time, or 800 hr of observation time.

Our detection limit would have been much lower if the length-to-width ratio had been a more accurate predictor of magnification, but future studies can target objects with known magnifications from more accurate lens models. We can use Figure 6 to determine the magnification required to detect sources of various mass with 25 hr integration time. For example, we should be able to detect a galaxy with M_Hi = 3.16 × 10⁹ out to z = 0.725 as long as it has μ > 30. A strongly lensed arc behind the cluster Abell 370 at z = 0.725 has been mentioned in Richard et al. (2010). It appears to be an SBc type galaxy, which typically have higher Hi mass (Roberts & Haynes 1994), with a total magnification of 32, and should be detectable with the GBT within 100 hr of observation time.

**Figure 6.** The magnification necessary to detect a galaxy of M_Hi = 10⁹ M_⊙ (green), 10^9.5 M_⊙ (red), 10¹⁰ M_⊙ (blue) at a given redshift using the GBT with a 25 hr effective integration time. The dotted line represents the integrated flux for each mass, the solid line indicates the 5σ detection limit with 25 hr effective integration time, and the dashed line indicates the magnification required to bring the integrated flux of the source up to the detection limit. The detection limit is calculated using the T_sys values of the GBT and fluctuates in accordance.
Download figure:
Standard image High-resolution image

5. CONCLUSIONS AND FUTURE WORK

Gravitational lensing has been used to measure emission from molecular gas and stars at higher redshift in the past and we suggest it can be used to measure Hi at high redshift with current telescopes. The lower frequencies of redshifted Hi have a denser RFI environment and, as our observations show, can lead to over 10% of the data being discarded due to wideband RFI. To recover as much information as possible from our data we created a custom sigma-clipping routine that removes wideband and narrowband interference. The narrowband interference is removed using sigma-clipping in the frequency domain, and wideband interference is removed using sigma-clipping in the Fourier domain. When we flag that data in the Fourier domain the information near the source is preserved and the wideband RFI removed, reducing the noise and flattening the baseline. Testing shows the routine is effective when the narrowband interference does not cover the signal, and when ≲10% of the integrations have wideband interference. We used this flagging technique on spectra from three sources behind the galaxy cluster Abell 773. Flagging the spectra with wideband interference in the Fourier domain rather than discarding them reduced the final noise in the spectra by about 25%. The final noise per channel of our spectra were 211 μJy at z = 0.487 and 204 μJy at z = 0.398. This flagging method can be combined with the magnification provided by gravitational lensing to detect Hi in individual galaxies out to and beyond a redshift of z = 0.5.

We have observed three sources for an effective integration time of 18.1 hr at z = 0.487 and 17.2 hr at z = 0.398. The sources have low magnification, and were not detected, but we were able to set a 3σ detection limit on their masses of M_Hi = 1.35 × 10¹⁰ M_⊙ and M_Hi = 6.36 × 10⁹ M_⊙, which is higher than their estimated masses of 4.7 × 10⁸ and 1.8 × 10⁹ M_⊙ respectively. In order to detect these sources we require 800 additional hours of observation time. Adding more observation time now seems costly, so we must identify other sources to observe that should be detectable based on their magnification.

We have used the criteria from Figure 6 to find other sources that are detectable with the GBT. Sources beyond z = 1 would require unrealistically large magnifications, effective integration times, or Hi masses to be detectable, so we have identified 28 lenses below that redshift that can be targeted for similar observations. We have already begun observations on two sources behind the cluster Abell 370, one with a magnification of ∼32 where a Hi detection should be possible in an effective integration time of 25 hr.

We thank the GBT staff for their assistance with the many observing sessions; Toney Minter and Ron Maddelena for their helpful discussions on flagging; and Johan Richard for providing us with unpublished magnification information and helpful discussions of their applicability. We also want to thank the reviewer for helpful comments that improved the paper. Some of this work was based on observations made with the NASA/ESA Hubble Space Telescope, obtained from the data archive at the Space Telescope Science Institute. STScl is operated by the Association of Universities for Research in Astronomy, Inc. under NASA contract NAS 5-26555. This work was partially supported by NSF CAREER grant AST-1149491, and NSF AAG grant AST-1412578.

THE SEARCH FOR Hi EMISSION AT z ≈ 0.4 IN GRAVITATIONALLY LENSED GALAXIES WITH THE GREEN BANK TELESCOPE

Article metrics

Permissions

Author e-mails

Author affiliations

ORCID iDs

Dates

ABSTRACT

1. INTRODUCTION