COMAP Early Science. VI. A First Look at the COMAP Galactic Plane Survey

Observations for the COMAP Galactic Plane Survey were made using the COMAP Pathfinder telescope (Lamb et al. 2022), sited in Owens Valley Radio Observatory (OVRO) in California. The telescope itself is of Cassegrain design and has an FWHM beamwidth of 45 consistent to ±4% across 26–34 GHz (Lamb et al. 2022). The Pathfinder has a focal plane array of 19 forward-facing pixels in a hexagonal pattern. Feed centers are separated by 65 mm giving a sky-angle offset between pixels of 1204 and allowing for 19 independent observations of the sky to be taken simultaneously.

The radio frequency signal from each feed is passed into a polarizer, which converts the left-circular wave into linear polarization accepted by a low-noise amplifier operating at 15–18 K. The outgoing noise wave from the amplifier has its polarization reversed on reflection by the secondary and therefore does not couple back to the amplifier to cause a ripple in the spectrum. The amplifier output is then downconverted in two stages. The first mixes the 26–34 GHz with a 24 GHz local oscillator (LO), producing the first intermediate-frequency signal at 2–10 GHz. This is split into two paths, one feeding a 2–6 GHz filter (band A), and the other a 6–10 GHz filter (band B). Band A is mixed with a 4 GHz LO to produce an in-phase (I) and a quadrature (Q) signal at 0–2 GHz. Each of these baseband signals is sampled in an 8 bit analog-to-digital converter and the I and Q signals are combined in a field-programmable gate array (FPGA) to produce the lower sideband (LSB) of the 4 GHz LO at 2–4 GHz, and the upper sideband (USB) at 4–6 GHz. Similarly, band B uses an 8 GHz LO to convert to the I and Q basebands, converted by an FPGA to produce the 6–8 GHz and 8–10 GHz sidebands. The four signals—LSB:A, USB:A, LSB:B, and USB:B—correspond to four continuous 2 GHz bands between 26 and 34 GHz on the sky.

The FPGAs then perform spectral analysis on 1024 channels for each 2 GHz band, with 2 MHz spectral resolution across the full 8 GHz bandwidth observable with COMAP. For a more complete description see Lamb et al. (2022). The resulting data set is collated alongside pointing, environmental, and housekeeping data and stored locally at Caltech.

The COMAP Galactic Plane Survey will cover the Galactic plane over the range 20° < ℓ < 220° in longitude and −2° < b < 2° in latitude. Observations started in 2019 June and are ongoing. Typically 1–2 hr per day of COMAP observing time is dedicated to the Galactic plane survey. Since 2019 June we have surveyed 20° < ℓ < 50° in Galactic longitude, totaling 834 hr of observing time. Observations are initially calibrated at the beginning and end of each observation using a thermal load with a known temperature (details in Foss et al. 2022) and then calibrated to an astronomical brightness scale using daily observations of the SNR Taurus A/Crab Nebula (Tau A; Section 2.4). All observations were taken during the day.

The survey was conducted by observing the Galactic plane in discrete patches, where each patch covers an area of approximately 4 deg². As the COMAP instrument focal plane spans approximately ∼1°, we nested neighboring patches to ensure a uniform sensitivity across the survey. To map out each patch we would begin scanning the telescope in the horizon frame with a Lissajous pattern, and allow the natural rotation of the sky to move the patch center through the field of view of the telescope. The Lissajous scanning strategy traces sinusoidal patterns in the azimuth and elevation that result in each pixel in the celestial frame being visited by many different scan paths, a condition that is required for the map-making method (Section 2.5) as it allows for the true sky signal and correlated noise to be separated.

The Lissajous pattern is defined by the pair of equations

$$\begin{eqnarray}&&{\delta }_{A}=\displaystyle \frac{R}{\cos (E)}\sin ({\omega }_{A}t+\phi ),\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{\delta }_{E}=R\sin ({\omega }_{E}t),\end{eqnarray} \tag{ 2 }$$

where δ_A and δ_E represent the offset in azimuth and elevation from the central azimuth (A) and elevation (E) coordinates, ω_E and ω_A are the angular velocities along each axis, R is the radius of the scans, and ϕ is the relative phase. We used R = 08 for the entire survey, the phase was alternated between $\pm \tfrac{\pi }{2}$, and the ratio of the angular velocities was randomly selected in the range 0.6 < ω_E /ω_A < 1. The telescope was driven close to its maximum rate in elevation (${V}_{E}^{\max }=0\buildrel{\circ}\over{.} 5\,{{\rm{s}}}^{-1}$) for all observations. We modulated the Lissajous pattern by changing the azimuth scan speed (${V}_{A}^{\max }=1^\circ \,{{\rm{s}}}^{-1}$).

The nominal on-sky system temperatures were measured to be in the range 30–44 K across the full 26–34 GHz COMAP band. We rejected a number of channels that had abnormally high system temperatures caused either by spectral aliasing at the edges of the bands or by resonances within the feed optics (see Lamb et al. 2022). After flagging bad channels, we averaged the native 2 MHz channels into wide 1 GHz bands.

When using the COMAP system for continuum science, the data are contaminated by substantial time-correlated noise (often referred to as 1/f noise; Harper et al. 2018) from two sources: fluctuations in the precipitable water vapor content in the atmosphere above the telescope, and small fluctuations in the gain or system temperature of the receiver low-noise amplifiers. Mitigating 1/f noise is critical to recovering the large-scale, diffuse Galactic structures and to achieving the lowest possible noise levels in the map. The strongest atmospheric 1/f noise can be mitigated by discarding data from days that have either poor or turbulent weather conditions. These can be determined by measuring the feed-to-feed noise correlation (as near-field atmospheric fluctuations will be strongly correlated between feeds), tracking the optical depth of the atmosphere using sky dips (Rohlfs et al. 2013), and measuring the power spectrum of the data to determine 1/f noise properties. We identified the very worst observations where the atmospheric fluctuations were several times the white noise of the receiver at timescales of 10 s or more and removed them, cutting the total optical depth (TOD) for all eight channels. This amounted to cutting approximately 17% of the observations, which were found to contain severe atmospheric contamination.

There is some contamination of the data from ground emission due to the local mountain ranges that lie to the east and west of OVRO. We characterized the ground emission profiles by performing 360° azimuth sweeps at fixed elevations. From these observations we could determine that the scale size of the ground emission is ≫ 1°, much larger than the typical patches observed. On the scale of a single scan (the period between two repointings, i.e., ≈1°; Foss et al. 2022) we approximated the ground emission (and other azimuth-correlated systematics) using a linear slope in azimuth, which gave a median amplitude for the ground emission across the survey of ≈6 mK. This ground emission slope was then subtracted from the TOD.

To suppress any remaining large-timescale 1/f noise fluctuations we used a running median filter with a scale size of 100 s (approximately equal to the typical time needed to complete a full scan—Section 2.2). Finally, we used the destriping map-making technique (see Section 2.5) in combination with the observing strategy to suppress 1/f noise down to timescales of 1 s.

Calibration of the COMAP Galactic Plane Survey is done in two steps. First we calibrate the data using a calibration vane, a remotely controlled microwave absorber acting as a thermal load that covers the feed array at the beginning and end of each observation (i.e., approximately every 40 minutes). We take the difference between the known vane temperature (T ≈ 290 K) and the cold sky (T ≈ 2.7 K) to estimate both the system temperature and the total gain of the receiver and antenna system. The vane calibration procedure is described in Foss et al. (2022).

The effect of atmospheric absorption at 30 GHz is significant. We track atmospheric opacities using sky dips, which involve slewing the telescope between elevations of 40° and 60° over a period of a few seconds. We find that the typical opacity of the atmosphere at OVRO is in the range τ ∼ 0.07–0.09, which equates to 15% absorption at 30° elevation. The vane calibration procedure corrects for atmospheric absorption due to the atmosphere along the line of sight. After the vane calibration we estimate the residual optical depth using the vane-calibrated observations of Tau A and Cassiopeia A (Cas A), and find the residual effect of atmospheric absorption to be δ τ ∼ 0.02 ± 0.01, which equates to a 2%–3% residual uncertainty in the calibration at the elevations used for the COMAP Galactic Plane Survey.

Absolute calibration to the main beam brightness scale is done using Tau A. We observe Tau A once per day and fit the peak brightness using a 2D Gaussian model. We derive the absolute calibration from the Tau A measurements by comparing them with the WMAP spectral fits and secular decrease models of Tau A (Tables 16 and 17 of Weiland et al. 2011). We do not apply color corrections when calculating model fluxes to calibrate on, as these are ≪1% in each 1 GHz band. We verify the absolute calibration using Cas A and Jupiter and find an overall accuracy of 3.2% with a relative calibration between bands of <1%. Models for Cas A are also taken from Weiland et al. (2011). The flux density model of Jupiter is derived by fitting a power law to the brightness temperature measurements taken using the CARMA instrument between 27 and 33 GHz (Karim et al. 2018), and using the ephemeris of Jupiter to track its solid angle on the sky. We combine these errors in quadrature giving an approximate 5% error on each of the eight maps in this work.

The COMAP beam has very good main beam isolation with the first sidelobe at more than 20 dB below the main beam response. Although this may be good enough in terms of confusion of nearby bright sources, these sidelobes still result in a scale-dependent calibration (see, e.g., Du et al. 2016, for a discussion of this issue). To estimate the level of the effect we radially integrate the COMAP beam models described in Lamb et al. (2022) out to the first null (i.e., the main beam) and third sidelobe (approximately 30' from the line of sight). We find that the integrated power changes by 10% between the main beam and $30^{\prime}$ (≈6× the beam FWHM). For these early results we are largely interested in sources that are either unresolved or only partially resolved by the main beam, so the effect of beam dependence in this work is minimal.

In order to suppress spurious large-scale contamination and any remaining 1/f noise within the COMAP continuum data we have implemented a bespoke destriping map-maker using the methods outlined in Delabrouille (1998), Sutton et al. (2009), and Sutton et al. (2010). Destriping map-making solves for 1/f noise by fitting linear offsets to the time-ordered data, but uses the scanning information to separate the 1/f noise from the true sky signal. To illustrate how destriping map-making works we will define the following data model:

$$\begin{eqnarray}&&{\boldsymbol{d}}={\boldsymbol{P}}{\boldsymbol{m}}+{\boldsymbol{F}}{\boldsymbol{a}}+{{\boldsymbol{n}}}_{w},\end{eqnarray} \tag{ 3 }$$

where d is a vector containing the time-ordered data (for a single band, i.e., one time stream), m is the true sky signal, P maps the sky signal to the time domain, a represents the offsets that describe the 1/f noise, F maps these offsets to d , and finally n _w is the uncorrelated white noise vector.

Solving for the offsets in Equation (3) results in

$$\begin{eqnarray}&&\hat{{\boldsymbol{a}}}={\left({{\boldsymbol{F}}}^{T}{{\boldsymbol{Z}}}^{T}{{\boldsymbol{n}}}^{-1}{\boldsymbol{Z}}{\boldsymbol{F}}\right)}^{-1}{{\boldsymbol{F}}}^{T}{{\boldsymbol{Z}}}^{T}{{\boldsymbol{n}}}^{-1}{\boldsymbol{Z}}{\boldsymbol{d}},\end{eqnarray} \tag{ 4 }$$

where $\hat{{\boldsymbol{a}}}$ is the maximum-likelihood estimate of the offsets, and ${\boldsymbol{n}}=\left\langle {{\boldsymbol{n}}}_{w}{{\boldsymbol{n}}}_{w}^{T}\right\rangle$ is the white noise covariance matrix. The matrix Z is defined as

$$\begin{eqnarray}&&{\boldsymbol{Z}}={\boldsymbol{I}}-{\boldsymbol{P}}{\left({{\boldsymbol{P}}}^{T}{{\boldsymbol{n}}}^{-1}{\boldsymbol{P}}\right)}^{-1}{{\boldsymbol{P}}}^{T}{{\boldsymbol{n}}}^{-1}.\end{eqnarray} \tag{ 5 }$$

The COMAP Galactic Plane Survey maps use a plane cylindrical polar projection in the Galactic coordinate frame. The nominal offset length used to destripe the data was 1 s, which corresponds to approximately 18' scales on the sky. Our destriping map-maker suppresses the 1/f noise by a factor of four or better on scales up to 30'. The average noise level of the final maps is 4.6 ± 1.2 mK arcmin².

The map-space processing was done using a custom-designed Python pipeline that operates in two steps: preparing and smoothing the maps, and then performing source extraction and photometry. This process began with any map-specific processes (such as reprojection or beam correction) and was then followed by two common final preparation steps in all maps.

In order to account for different beam sizes the data were smoothed to a common resolution—in this analysis, 5'. The units of each map were also converted to a common unit (MJy sr⁻¹) to allow easy extraction of integrated flux densities, and to allow maps to be readily compared side by side.

3.2.1. Low-frequency Data Sets

3.2.2. High-frequency Data Sets

The first task required identifying bright, relatively compact sources within a small sample region of the COMAP survey for early analysis. We focused on sources that can be measured easily due to their brightness relative to their background, and their compact size relative to that of the beam, i.e., ≲5'. To do this the COMAP 30 GHz map was put through the preparation described in 3.1 and then through a Python implementation of Source Extractor Python (SEP; Barbary 2016), a Python wrapper for SExtractor (Bertin & Arnouts 1996).

Within the SEP implementation several options were invoked to ensure a reliable source list was returned which minimized the number of false detections. The first process used was filtering—this was done with an unnormalized Ricker wavelet of the form

$$\begin{eqnarray}&&{f}_{\mathrm{MH}}(r,\sigma )=\left(1-\frac{1}{2}\frac{{r}^{2}}{{\sigma }^{2}}\right)\exp \left\{-\frac{1}{2}\frac{{r}^{2}}{{\sigma }^{2}}\right\},\end{eqnarray} \tag{ 6 }$$

where r represents the radius from the center of the filter, and σ represents the filter's standard deviation. The filter was computed on a 15 × 15 array (which equates to a ${15}^{{\prime} }\times {15}^{{\prime} }$ space when projected onto the map) with a standard deviation of 22 to match the COMAP beam.

When inputting a signal-to-noise cutoff into SEP, we found that the influence of the diffuse background and the high density of sources caused the noise estimates to be greater than those measured by using an aperture on the background by an order of magnitude. As such, we input a measured background noise level of 6.5 ± 2.4 mK on scales of $15^{\prime}$ and applied a 5σ detection limit on all sources.

After this analysis of the raw COMAP 30 GHz map 21 candidate sources were identified. Through checking each source by eye in all eight COMAP bands and the 5 GHz Parkes and Planck maps, sources were either verified or rejected. These sources were verified by a clear detection in all eight COMAP bands in addition to a clear detection in one of the two other maps (verifying a strong dust and/or free–free component from the source). However, performing photometry on some of these remaining sources was still nontrivial. We therefore further removed additional sources that were still weak and if the background was sufficiently bright/complicated those whose flux densities were not robust to variation in the background annulus locations. This left a total of nine sources with robust SEDs to present in Section 5.4.

We used aperture photometry to measure the flux densities of the sources identified in the previous section. While we see that Gaussian fitting may be more suited to such a complex region as the Galactic plane, particularly for sources embedded within extended emission, we implement a simple aperture photometry technique here to provide an early indication of the spectral composition of such sources on small scales. Since we are focused on brighter sources, aperture photometry should provide a reliable estimate of the flux density.

For the H ii regions, we used a constant aperture radius of $r={5}^{{\prime} }(1\sigma )$ to integrate the flux of each source. We found this to be suitable given that each source on the initial extracted list had a fitted standard deviation in the range $2\buildrel{\,\prime}\over{.} 2\lt \sigma \lt 2\buildrel{\,\prime}\over{.} 5$, giving a ≈2σ aperture radius size. For background measurements we used an annulus drawn between 67 and $8\buildrel{\,\prime}\over{.} 3$ from the source coordinates and used the median pixel value to estimate a background that was robust against nearby bright sources.

In the case of SNRs, we treated these separately due to the large amount of variability between sources, and the tendency of SNRs to appear as extended sources. For these sources we did not perform a specific source extraction, but used the values given in Green (2019) to set initial values for the radius and central coordinates of the aperture and annulus used, adjusting these to fit the SNRs as they appeared in the maps and the background. We then used the adapted central coordinates and radii in the aperture photometry pipeline in Section 4.2 to obtain flux measurements from our maps.

We fit the flux density of each H ii source measured using aperture photometry (Section 4.2) with a three-component model of the free–free (S_ff), thermal dust (S_td), and AME components (S_AME):

$$\begin{eqnarray}&&S={S}_{\mathrm{ff}}+{S}_{\mathrm{td}}+{S}_{\mathrm{AME}}.\end{eqnarray} \tag{ 7 }$$

We modeled the free–free emission flux density by first calculating the free–free brightness temperature and converting it to flux density units via

$$\begin{eqnarray}&&{S}_{\mathrm{ff}}=\frac{2{k}_{{\rm{B}}}{T}_{\mathrm{ff}}{{\rm{\Omega }}}_{\mathrm{beam}}{\nu }^{2}}{{c}^{2}},\end{eqnarray} \tag{ 8 }$$

where k_B represents the Boltzmann constant, Ω_beam the beam solid angle, ν the frequency, and c the speed of light. T_ff is defined as (e.g., Draine 2011)

$$\begin{eqnarray}&&{T}_{\mathrm{ff}}={T}_{e}(1-{e}^{-{\tau }_{\mathrm{ff}}}),\end{eqnarray} \tag{ 9 }$$

where T_e is the the electron temperature and τ_ff is the free–free emission optical depth:

$$\begin{eqnarray}&&{\tau }_{\mathrm{ff}}=5.468\times {10}^{-2}{T}_{e}^{-1.5}{\nu }^{-2}\mathrm{EM}{g}_{\mathrm{ff}}\,,\end{eqnarray} \tag{ 10 }$$

where g_ff is the Gaunt factor, which for these frequencies can be approximated as

$$\begin{eqnarray}&&{g}_{\mathrm{ff}}=\mathrm{ln}\left(\exp \left[5.960-\displaystyle \frac{\sqrt{3}}{\pi }\mathrm{ln}({\nu }_{\mathrm{GHz}}{T}_{4}^{-1.5})\right]+2.71828\right),\end{eqnarray} \tag{ 11 }$$

where T₄ = T_e × 10⁻⁴. For the SED fitting we let the emission measure be a free parameter, while the electron temperature we fixed to a typical value of T_e = 7500 K (e.g., Paladini et al. 2003). However, the fit was only weakly dependent on the choice of T_e .

The thermal dust emission was modeled as a modified blackbody curve given by

$$\begin{eqnarray}&&{S}_{\mathrm{td}}=2h\frac{{\nu }^{3}}{{c}^{2}}\frac{1}{{e}^{h\nu /{k}_{{\rm{B}}}{T}_{d}}-1}{\left(\frac{\nu }{353\,\mathrm{GHz}}\right)}^{{\beta }_{d}}{\tau }_{353}{{\rm{\Omega }}}_{\mathrm{beam}}\,,\end{eqnarray} \tag{ 12 }$$

and we fit for the optical depth of thermal emission at 353 GHz (τ₃₅₃), the dust temperature (T _d ), and the emissivity spectral index (β _d ). It is worth noting that there are well-known degeneracies between the emissivity spectral index and dust temperature (e.g., Dupac et al. 2003; Désert et al. 2008; Planck Collaboration et al. 2014).

We modeled the AME spectrum as a lognormal curve as this is a flexible model that is a good approximation to a spinning dust spectrum (Bonaldi et al. 2007; Stevenson 2014; Dickinson et al. 2018); it is defined as

$$\begin{eqnarray}&&{S}_{\mathrm{AME}}={A}_{\mathrm{AME}}\exp \left\{-\displaystyle \frac{1}{2}{\left(\displaystyle \frac{\mathrm{ln}(\nu )-\mathrm{ln}({\nu }_{\mathrm{AME}})}{{w}_{\mathrm{AME}}}\right)}^{2}\right\}\,,\end{eqnarray} \tag{ 13 }$$

where we fit for the amplitude of the AME (A_AME), the peak frequency (ν_AME), and the width of the lognormal (w_AME).

For the SNRs we did not fit for a free–free or AME component but instead fit for synchrotron emission (S_sync) so the SED model became

$$\begin{eqnarray}&&S={S}_{\mathrm{sync}}+{S}_{\mathrm{td}},\end{eqnarray} \tag{ 14 }$$

where the synchrotron component is modeled as a simple power law:

$$\begin{eqnarray}&&{S}_{\mathrm{sync}}={A}_{\mathrm{sync}}{\nu }^{\alpha },\end{eqnarray} \tag{ 15 }$$

where we fit for the spectral index of the synchrotron (α), and the synchrotron amplitude at 1 GHz (A_sync).

The SEDs were fitted using an MCMC analysis, and the emcee ensemble sampler (Goodman & Weare 2010; Foreman-Mackey et al. 2013a) as the backend. The implementation we used is the same as that described in Cepeda-Arroita et al. (2021) with a modification to account for the correlation in noise between the COMAP bands. For each fit we used 300 chains, each with 5000 steps. The burn-in time for each fit required between 1300 and 1400 steps as assessed by eye, and so we discarded the first 1500 from each chain. The typical correlation length of the samples was 390 steps, which we used to thin the chains and to suppress sample-to-sample correlations. We checked each chain for convergence, and discarded those that failed.

At the resolution of COMAP there are no surveys that cover the frequency range 40–100 GHz. The lack of data at these frequencies means that the peak and width of the AME lognormal are poorly constrained. To improve the convergence of the MCMC fitter we implemented Gaussian priors on ν_AME and w_AME. The Gaussian priors were informed by measuring the integrated flux density of W43 using the same 1° radius aperture and background annulus (13–17) used in Irfan et al. (2015) and Génova-Santos et al. (2017). By using maps smoothed to $30^{\prime}$ resolution we were able to use the data sets in Table 2 alongside the lower-resolution Planck Low Frequency Instrument (LFI) 28.4, 44, and 70 GHz bands. Fitting for a model with free–free emission, AME, and thermal dust emission we constrained a mean peak frequency and lognormal width for the region. The final Gaussian priors we used for the H ii SED fits were N(28 ± 15 GHz) for the peak frequency (ν_AME), and N(0.6 ± 0.2) for the width (w_AME).

In Figure 1 we show the COMAP 30.5 GHz map covering 20° < ℓ < 40° on the top panel, and the total number of integrations per ${1}^{{\prime} }$ pixel in the bottom panel. The map units are in millikelvins and the map has a resolution of $4\buildrel{\,\prime}\over{.} 5$. We detect with a high signal-to-noise ratio (S/N) several bright giant molecular cloud regions such as G023.3−00.3 (e.g., Messineo et al. 2014) and W43 at (ℓ, b) = (254, −02) (e.g., Nguyen Luong et al. 2011); several giant H ii regions such as G24.5−0.0 and W42 at (ℓ, b) = (254, −02), as well as many known H ii regions and H ii complexes (e.g., Paladini et al. 2003; Anderson et al. 2014); and also several SNRs (Section 5.5). Besides discrete sources we detect extended emission in the form of diffuse structures and spurs. Most of the diffuse emission is caused by diffuse ionized gas formed from the leakage of ionizing radiation from nearby O/B stars into the ISM (Zurita et al. 2000), but there will also be dust-correlated AME due to spinning dust grains (Hensley et al. 2016; Dickinson et al. 2018) potentially contributing up to ≈20%–45% of the observed brightness (Planck Collaboration et al. 2011, 2015).

Zoom In Zoom Out Reset image size

The map has a high S/N across much of the Galactic plane. The noise level of the map varies between 2 and 3 mK beam⁻¹ away from the edges, or equivalently 0.1–0.15 Jy beam⁻¹. Due to the filtering of the time-ordered data on timescales corresponding to the typical scan length of ∼1°, the largest scales are not well constrained. We find that there is a loss of flux of 15% or more on scales larger than about 30'. For example, a comparison between the integrated flux density of W43 at 30' resolution over a 2° diameter aperture in the COMAP 28.5 GHz data and that measured by the Planck LFI at 28.4 GHz (504 ± 26 Jy; Irfan et al. 2015) shows that we are underestimating the integrated flux density by ≈15%. However, as discussed in Section 2.4 the data have been calibrated to 5% or better at the main beam scale (taking into account both opacity errors and those on the raw calibration) out to a few times the FWHM (∼12'); therefore we will primarily focus on discrete (i.e., clear, high-S/N) sources for this first analysis. In future releases we will use the COMAP beam models (Lamb et al. 2022) to deconvolve the map and use simulations to help improve the data processing in order to preserve large-scale structures.

In the following sections we will discuss several initial analyses of the COMAP Galactic Plane Survey. In Section 5.2 we will compare the pixel brightnesses of the diffuse emission at 30 GHz with surveys at 5 and 353 GHz. In Section 5.3 we discuss the general contribution of UCHii regions to the total emission observed at COMAP frequencies using data from the CORNISH survey UCHii catalog. In Section 5.4 we discuss nine H ii regions selected from the COMAP Galactic Plane Survey and look for evidence of AME. In Section 5.5 we look at the fitted SEDs of six known supernovae (SNRs) over the full 20° < ℓ < 40° range shown in Figure 1. Finally, in Section 5.6 we present preliminary detections of RRLs from both compact and diffuse ionized gas within the COMAP band within the area marked W43 in Figure 2.

Zoom In Zoom Out Reset image size

We begin the analysis with an initial look at the correlation of the diffuse emission seen at 30 GHz shown in Figure 1 with the 5 GHz Parkes map (primarily tracing free–free emission; Calabretta et al. 2014) and the 353 GHz Planck map (tracing dust emission; Planck Collaboration et al. 2020). We will derive the average spectral index and Pearson correlation coefficient of the diffuse emission between each frequency pair. If AME is present in the diffuse emission at 30 GHz we will expect to see a rising spectrum between 5 and 30 GHz that is greater than that predicted for free–free emission (α_5–30 > −0.1), and significant correlation between the 30 and 353 GHz data. We account for the possible filtering on large scales discussed earlier by including a conservative 15% calibration uncertainty on the COMAP data (due to the loss of flux density on large scales discussed in Section 5.1).

Before comparing the pixel brightnesses we first smooth all the data sets to a common resolution of 5' and change the pixel size to match the resolution of the maps to reduce pixel-to-pixel correlations in the noise. We then mask all pixels with a pixel brightness of less than 1 MJy sr⁻¹ at 5 GHz; the mask area is shown by the lowest contour in Figure 1. We also mask the very brightest sources that exceed 6 MJy sr⁻¹ at 30 GHz. The mask ensures we have good S/N in the majority of pixels and mitigates the possibility of including low-level large-scale modes that are not well constrained away from the Galactic plane.

In Figure 3 we compare the pixel brightness in the 5 GHz and 353 GHz maps with COMAP 30 GHz data. We can see there is a strong correlation between the 30 GHz data and both 5 GHz and 353 GHz emission. The Pearson correlation coefficient between each frequency pair is consistent: r_5–30 = 0.72, r_30–353 = 0.68, and r_5–353 = 0.70. Strong correlations between all three bands are not unexpected near the Galactic plane as the complex dust and gas dynamics of these regions leads to significant mixing of ISM phases.

Zoom In Zoom Out Reset image size

The dashed lines in Figure 3 show the best-fit linear relationships between the pixel brightnesses. We convert the fitted gradients into the spectral index between pairs of frequencies by

$$\begin{eqnarray}&&\alpha =\frac{\mathrm{log}\left(d{S}_{{\nu }_{0}}/d{S}_{{\nu }_{1}}\right)}{\mathrm{log}\left({\nu }_{0}/{\nu }_{1}\right)},\end{eqnarray} \tag{ 16 }$$

where α is the spectral index between frequencies ν₀ and ν₁.

The spectral index that we measure between 5 GHz and 30 GHz is α_5–30 = 0.035 ± 0.017, which is inconsistent with the expected free–free spectral index of α_ff ≈ −0.1 (Draine 2011) at the level of 7.9σ. It is possible for a rising spectrum between 5 and 30 GHz to be due to optically thick free–free emission from embedded UCHii regions, but as discussed in Section 5.3, we find that on average UCHii regions have a negligible impact on these angular scales at 30 GHz. We therefore interpret this excess as due to AME, probably in the form of spinning dust. We can estimate what fraction of the emission measured at 30 GHz is due to AME by

$$\begin{eqnarray}&&{\eta }_{\mathrm{AME}}=1-\displaystyle \frac{{\left({\nu }_{1}/{\nu }_{0}\right)}^{-0.1}}{{\left({\nu }_{1}/{\nu }_{0}\right)}^{{\alpha }_{5\mbox{--}30}}},\end{eqnarray} \tag{ 17 }$$

where ν₁ = 30.5 GHz, and ν₀ = 5 GHz. We find a fractional AME of η_AME = (22 ± 2)%. There have been several previous estimates of the AME fraction around 30 GHz. Planck Collaboration et al. (2011), using a template fitting method, estimated the AME fraction at 28.4 GHz to be (25 ± 5)% across all longitudes, but found particular longitudes to have as high as 100%; however such high AME fractions are likely due to modeling errors in complex regions. Planck Collaboration et al. (2015) also performed a template fitting analysis but made use of RRL data over the range 5° < ℓ < 60° and found an average AME fraction of (45 ± 1)%. Planck Collaboration et al. (2018) found the average AME fraction of individual bright AME regions is (50 ± 2)%.

In future work we will expand this analysis to a full correlation analysis where we jointly fit for emission components at 30 GHz using templates in a similar way to what has been done at lower resolutions (e.g., Davies et al. 2006). However, since the free–free emission and thermal dust emission are so strongly correlated (r_5–353 = 0.70) such an analysis will require careful consideration of this correlation.

As described in Section 4.3, the SED fitting assumes that free–free emission may be fit by a single optically thin and spatially smooth component. Therefore any contribution from younger, compact H ii regions with rising spectra and higher turnover frequencies may erroneously be classified as AME. These young compact H ii regions can be further classified as UCHii and hypercompact H ii (HCHii) regions depending on their electron density and emission measure (e.g., Churchwell 2002; Kurtz 2002). The most compact objects (UCHii/HCHii) can exhibit turnover at frequencies of tens of gigahertz (Kurtz 2002), although the flux density of individual objects will typically be low (∼mJy). In this section we describe the average contribution of UCHii regions to the emission at 30 GHz.

To account for these contributions, one would ideally like to survey at high frequency (e.g., 30 GHz) and high angular resolution (∼arcsec) to allow each source to be accounted for (each source will typically be quite weak—typically at the level of tens to hundreds of millijanskys). Unfortunately while COMAP does have the required frequency coverage, it does not have the angular resolution necessary to perform such a search; however we may consult external catalogs to search for compact sources that may influence our findings. The most appropriate survey to date is the 5 GHz CORNISH VLA radio survey (Hoare et al. 2012) of the northern Galactic plane. We use the UCHii catalog from the 5 GHz CORNISH survey (Purcell et al. 2013; Kalcheva et al. 2018). The CORNISH survey catalogs ∼3000 sources in total at 5 GHz with 15 resolution, 54 of which lie within our survey area of 20° < ℓ < 40°. Considering that we report 30 GHz excess fluxes of ≈1 Jy, we would typically require ∼100 UCHii regions to account for the excess flux observed.

The catalog is estimated to be 90% complete to point sources at 3.9 mJy but is less complete for extended emission, resolving out sources larger than 14''. However, these extended H ii regions will, by definition, have lower densities and therefore lower turnover frequencies, i.e., they will not be UCHii regions and are therefore less likely to contribute to the excess emission at ∼30 GHz.

All the sources in the CORNISH UCHii catalog were classified based on their morphological and positional similarity to IR counterparts in the 8 μm band of the GLIMPSE survey (Kalcheva et al. 2018; GLIMPSE Team 2020). Proximity to IR clusters and dust lanes was used to distinguish UCHii regions from other objects meeting the above criteria such as massive young stellar objects and planetary nebulae.

We have assessed the contribution from the 54 cataloged UCHii sources to the fitted AME and found it to be negligible. However, we must also consider the potential for sources that were not included in the catalog to make a significant contribution. These sources fall into two main categories: (i) any compact H ii sources that appeared in the parent CORNISH catalog but were not detected in GLIMPSE (referred to as "IR-quiet") and so could not be classified as UCHii regions, and (ii) sources falling below the completeness level of the parent CORNISH catalog.

Of the CORNISH IR-quiet sources 80% have flux densities of ≤13 mJy. Therefore, we conservatively estimate the contribution of sources missing from the catalog by assuming a true completeness level of the UCHii catalog of 13 mJy (over three times the point-source completeness limit of the CORNISH survey).

Yang et al. (2021) observed 116 young H ii regions that have rising spectra from 1 to 5 GHz (5 in our survey area), and found 20 sources with turnover frequencies >5 GHz (4 in our survey area), with maximum and mean turnover frequencies of 16.7 GHz and 9.7 GHz, respectively. If we consider a source with a 5 GHz flux density of 13 mJy and assume the source to have the maximum turnover frequency (ν_t = 16.7 GHz) reported by Yang et al. (2021), its 30 GHz flux density is just ∼13% of the mean excess (ΔS₃₀ in Table 3). This contribution decreases significantly to ∼4% if we assume the source to have the mean turnover frequency (ν_t = 9.7 GHz).

Table 3. Fitted SED Parameters for H ii Sources Analyzed with Their Associated Geometric Mean Widths from Source Extraction (θ) and In-band Spectral Indices (α₂₆₋₃₄)

Source		In-band	Thermal Dust			Free–Free	AME Lognormal
Name	θ	α26−34	Td	β	${\mathrm{log}}_{10}({\tau }_{353})$	EM × 10−3	AAME	νAME	wAME	ΔS30	ν × 1018
	(')		(K)			(pc cm−6)	(Jy)	(GHz)		(Jy)	(Jy sr–1 cm2 H–1)
G027.488+00.199	4.5	0.5±0.1	30±3	1.50±0.24	−3.79±0.07	12.7±0.4	...	...	...	<0.13	<1.25
G030.744−00.036	6.7	−0.0±0.1	22±1	2.10±0.08	−2.19±0.02	398±14	...	...	...	<1.49	<0.5
G031.125+00.294	3.9	−0.8±0.5	27±4	1.30±0.29	−3.71±0.13	5.3±0.4	...	...	...	<0.21	<2.8
G027.266+00.165	5.9	0.8±0.1	22±1	2.02±0.16	−3.56±0.05	7.1±0.7	1.0±0.5	41±17	0.81±0.21	0.98±0.10	10.1±3.8
G028.289−00.356	3.8	0.1±0.1	23±1	2.03±0.16	−3.40±0.05	5.7±0.6	0.6±0.4	40±16	0.73±0.23	0.52±0.07	9.2±3.5
G028.600+00.035	6.6	0.4±0.1	23±1	1.79±0.12	−3.14±0.03	14.0±1.0	1.2±0.3	44±13	0.71±0.20	1.06±0.11	3.3±1.2
G029.038−00.636	12.1	0.4±0.1	27±2	1.70±0.19	−3.35±0.05	17.5±1.2	3.6±1.0	47±10	0.56±0.13	1.97±0.14	3.0±1.1
G030.443+00.435	6.1	1.0±0.2	22±2	1.66±0.24	−3.34±0.06	5.2±0.5	0.9±0.3	47±13	0.59±0.20	0.70±0.08	3.2±1.2
G030.508−00.447	4.9	0.4±0.1	23±2	1.73±0.23	−3.35±0.05	6.9±0.8	1.1±0.4	45±14	0.66±0.22	0.89±0.10	8.3±3.1

Note. The fitted parameters are given for the models discussed in Section 4; the model excess emission (ΔS₃₀) and emissivity at 30 GHz (_ν ) are calculated from the excess and optical depth at 353 GHz. For non-AME sources we present 2σ upper limits on the last two parameters: 30 GHz excess and emissivity.

Download table as:  ASCIITypeset image

Given the rarity of UCHii regions with such high turnover frequencies, we conclude that sources missing from the CORNISH UCHii catalog do not make a significant contribution to the reported AME in this paper. The largest contribution in any of the sources initially found by our analysis is seen to be ≈4% although most sit well below this contribution. Nevertheless, as we expand the COMAP survey, there might be some lines of sight where UCHii regions, particularly if there is a cluster within a molecular cloud, could account for a major fraction of the total flux density. In Section 5.4 we also provide details on the UCHii region contributions to each of the six AME sources.

At frequencies of ν < 100 GHz H ii regions are dominated by free–free emission, which has a well-defined spectral shape (Draine 2011) that at frequencies within 5 < ν < 100 GHz is often optically thin with a spectral index of α_ff ≈ −0.1. There are several examples of AME being detected in H ii regions (Watson et al. 2005; Dickinson et al. 2007; Planck Collaboration et al. 2018), as well as observations associating the AME with the swept-up circumstellar material at the boundaries of H ii regions (Tibbs et al. 2010, 2012; Cepeda-Arroita et al. 2021). However, not all H ii regions show evidence of AME (Scaife et al. 2007, 2008). In this section we will discuss nine sources selected from the COMAP Galactic Plane Survey that are coincident with H ii sources, six of which we find to have a >3σ 30 GHz excess that is indicative of AME, and three of which show no evidence of AME.

In Figure 4 we show the best-fit SED models to the non-AME sources that are dominated by optically thin free–free emission at 30 GHz, and Figure 5 shows the model fits to the six sources that exhibit a significant excess at 30 GHz above the fitted free–free emission component.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In Table 4 we give the χ²/dof of each region, as well as the relative contribution to the χ²/dof for data between 2.7–10 GHz (${\chi }_{\mathrm{lo}}^{2}$), 26–35 GHz (${\chi }_{\mathrm{COMAP}}^{2}$), and 100–4000 GHz (${\chi }_{\mathrm{hi}}^{2}$). For most regions we find 0.5 < χ²/dof < 1.5. Estimates where the χ²/dof is less than unity are due to background correlation uncertainties not being accounted for and to the correlation between the calibration of each survey. The assumption that the calibration uncertainties we give in Table 2 are uncorrelated, results in an overestimate of the uncertainty for some regions. The worst-fitting regions are G030.744−00.036 (the core of W43) and G029.038−00.636 (RCW 175) with χ²/dof ∼ 3. For G030.744−00.036 the large χ²/dof is driven by the low-frequency data; specifically the flux density measured at 5 GHz using the Parkes survey is underestimated relative to the two other low-frequency surveys. We suppose that this is due to issues with the calibration of the Parkes data in this region, likely due to significant sidelobe pickup from the nearby, bright diffuse emission that surrounds G030.744−00.036 (although further investigation would be required to prove this). The other region with a high χ²/dof is G029.038−00.636; in this case we find that the issue is associated with just the Akari data. A comparison between the IRAS and Akari data does not reveal any clear morphological differences (e.g., no point sources have been subtracted from the reprocessed IRAS data in the region), suggesting a systematic calibration issue with the Akari data along this line of sight.

Table 4. Summary of Reduced χ²/dof Values Determined for the Nine Sources in Table 3

Name	χ2/dof	${\chi }_{\mathrm{lo}}^{2}$	${\chi }_{\mathrm{COMAP}}^{2}$	${\chi }_{\mathrm{hi}}^{2}$
G027.266+00.165	1.53	0.18	0.97	0.39
G030.744−00.036	3.42	1.9	0.46	1.06
G031.125−00.294	1.01	0.34	0.38	0.29
G027.488+00.199	0.53	0.02	0.34	0.17
G028.289−00.356	0.55	0.04	0.34	0.17
G028.600+00.035	0.61	0.08	0.15	0.38
G029.038−00.636	2.99	0.59	0.08	2.33
G030.443+00.435	1.31	0.38	0.25	0.68
G030.508−00.447	1.52	0.11	0.11	1.3

Note. χ²/dof values are given for the whole fitted data set alongside the contributions from data points with ν < 26 GHz (${\chi }_{\mathrm{lo}}^{2}$), for the eight COMAP bands (${\chi }_{\mathrm{COMAP}}^{2}$), and for data sets with ν > 34 GHz (${\chi }_{\mathrm{hi}}^{2}$).

Download table as:  ASCIITypeset image

We estimate the emission measure ($\mathrm{EM}\equiv \int {n}_{e}^{2}dl$) of each source from the fitted amplitude of the free–free model described in Section 4. We find that the aperture-averaged EM of the sources in our sample is in the range 5000–20,000 pc cm⁻⁶, which is typical of classical H ii regions (Kurtz 2002). The exception is W43 itself, which has a higher EM of 400,000 pc cm⁻⁶, comparable to previous estimates of the brightest (and densest) central source with EM = 720,000 pc cm⁻⁶ (Downes et al. 1980), and as expected given that it is a massive star formation region. As all of the sources are either approximately the size of the COMAP beam or slightly extended we are not significantly underestimating the true EM of the sources due to beam dilution; however the value does represent an average over the measured aperture area.

We also fit models for thermal dust emission, which give dust temperatures (T_d in Equation (12)) in the range 23–30 K. These measurements agree with Anderson et al. (2012), who reported dust temperatures of 25.3 ± 2.4 K averaged across whole H ii regions. Notably these dust temperatures are higher than those reported by Planck Collaboration et al. (2016b; 19.4 ± 1.3 K), who mainly considered the high-latitude sky, where dust is known to be colder.

Besides fitting full SED models we fit for the spectral index of the emission measured within the COMAP frequency band (26–34 GHz). The in-band spectral indices are given in Table 3 (α_26–34). We find that there may be evidence that AME regions are more likely to have rising spectra within the COMAP band; this will become clearer in the future as we measure more AME and non-AME regions in the COMAP Galactic Plane Survey.

The peak frequencies of the candidate AME sources are given in Table 3. We find that our peak frequencies are higher than the 25–30 GHz range found previously (e.g., Planck Collaboration et al. 2016a, 2018), with a mean value of ν_AME ∼ 40 GHz. We test several different priors and find that changes in the peak frequency for different priors are all consistent within the uncertainties. The apparently higher than expected AME peak frequency must therefore be driven by the lack of data between 40 and 100 GHz, meaning we cannot constrain the downturn in the AME spectrum. Although the peak frequencies we measure are generally high, they are still consistent within 1σ with the expected AME peak frequency of ∼30 GHz.

The AME widths that we measure are within 0.55 < w_AME < 0.85, similar to the range of widths measured in the λ Orionis ring (Cepeda-Arroita et al. 2021). In general, we find that the AME width is not well constrained for these sources for the same reason we cannot easily constrain the peak frequency (i.e., no data between 40 and 100 GHz). However, we do notice that there tends to be a relationship between the width of the AME spectrum and the measured in-band spectral index, with steeper in-band spectra leading to narrower widths implying we do have some constraining power from the COMAP data alone. However, at present this is not a clear relationship; for example G027.27.266+00.165 has a steep in-band spectrum but also a wide w_AME, which is mostly driven by an excess of emission seen in the 10 GHz Nobeyama data.

In order to obtain the AME emissivity (brightness per unit column density of dust) for each source (_ν ) we use the excesses and τ₃₅₃ values reported in Table 3. We convert τ₃₅₃ to a column density via (8.4 ± 3.0) × 10⁻²⁷ cm² H^–1 as reported by Planck Collaboration et al. (2014) for the whole sky. The uncertainties are dominated by this conversion factor. These values make up the last column in Table 3 and show a clear difference between non-AME sources (most of which are consistent with zero emissivity) and the AME regions, which typically have emissivities of the order of 10⁻¹⁷ at 30 GHz.

In Figure 4 we show the SEDs of the H ii regions in which we find no evidence of AME. The SEDs are fitted with a two-component free–free plus thermal dust model. The region G027.488+00.199 is located near to the H ii region G027.266+0.165 and just outside of the ring structure marked in Figure 2. The H ii region G031.125+00.294 lies near the exterior of the larger W43 complex. Both G027.488+00.199 and G031.125+00.294 are classical H ii regions; they contain a number of infrared dark clouds (IRDCs; Peretto & Fuller 2009) but are not associated with any known stellar clusters or ionizing main-sequence stars (Reed 2003; Bica et al. 2019). The region G030.744−00.036 is centered on W43, one of the Galaxy's most active star formation regions (e.g., Nguyen Luong et al. 2011); using a central 8' aperture we measure a flux density of ∼100 Jy, approximately 20% of the flux density of the entire region (Irfan et al. 2015; Génova-Santos et al. 2017).

In the following we will discuss the properties of each extracted H ii region in more detail. In Figure 6 we show the Spitzer 8 μm GLIMPSE (image; Churchwell et al. 2009; GLIMPSE Team 2020) and Herschel 250 μm Hi-GAL survey (contours; Molinari et al. 2016) data of each region in which we detect AME. All of these regions are sites of active or recent star formation. The annotations are the H ii region designations defined in Anderson et al. (2014); we also mark the locations of bright 5 GHz sources from the CORNISH survey (Purcell et al. 2013), the locations of IRDCs (Peretto & Fuller 2009), and any known O/B stars (Reed 2003).

Zoom In Zoom Out Reset image size

5.4.1. G027.266+00.165

5.4.2. G028.289−00.356

5.4.3. G028.600+00.035

5.4.4. G029.038−00.636

5.4.5. G030.443+0.435

5.4.6. G030.508−00.447

Few SNRs are readily detectable at higher radio frequencies (ν ≳ 30 GHz), owing to their synchrotron spectra's typical behavior of steeply falling with frequency (α ∼ −0.7) and to increased background. However, the brightest SNRs and those with a flatter spectrum (α > −0.5) are detectable. Those with flat radio spectra tend to have a filled-center or mixed morphology (rather than just a shell) and are often associated with a pulsar wind nebula (PWN); the Crab Nebula is one of the best examples (for a review, see Dubner & Giacani 2015).

A wide range of spectra have been observed, particularly over a wide frequency range, where a different population of electron energies are probed. The differences are thought to be due to the various intrinsic and extrinsic sources of energy injection/losses, resulting in a variety of spectra and morphologies (Urošević 2014; Dubner & Giacani 2015). Thus, high-frequency observations are useful for understanding energy losses/reacceleration as well as for characterizing nonsynchrotron components and distinguishing them from free–free emission and AME. Indeed, a number of SNRs have been reclassified as H ii (e.g., Dokara et al. 2021).

We have inspected the entire 30 GHz COMAP survey to date at the location of the 33 known SNRs from the most recent version ¹⁶ of the Green supernova catalog (Green 2019) over the longitude range l = 20°–40°. Most of these SNRs show no obvious source in the COMAP map, or are confused with other nearby sources and background emission. There are six SNR sources of interest, three of which are very extended (≳10') and three are compact or only slightly larger than the beamwidth (≳5'). We also look at the positions of new SNR candidates from Dokara et al. (2021) but no obvious detection is possible.

We briefly discuss them in turn and include our flux density measurement at 30 GHz along with the spectral index from 2.7 GHz to 30 GHz and the in-band spectral index fitted separately over the 26–34 GHz COMAP frequency range. The results are given in Table 5. For each SNR, we adjust the aperture size to be suitable for the SNR and use a background annulus that is 1.3–1.6 times the radius. The SEDs of these six SNRs are presented in Figure 7, showing our measurements from the maps available to us as well as values from the literature. Note that in order to stay as consistent with the literature as possible, we use the name (and coordinate convention) first allocated to each of the six SNRs when referring to it.

Zoom In Zoom Out Reset image size

5.5.1. G34.7−0.4 (W44)

5.5.2. G021.5−0.9

5.5.3. G021.8−0.6 (Kes 69)

5.5.4. G31.9+0.0 (3C 391)

5.5.5. G35.6−0.4

5.5.6. G39.2−0.3 (3C 396)

The ionized gas within H ii regions and the surrounding diffuse interstellar gas (DIG) within the warm ionized medium (WIM) not only emit continuum free–free emission but also, when the gas is optically thin, can be observed via RRLs (Kardashev 1959). Besides being an alternative probe of the WIM, RRLs provide additional velocity information, and can be used to study the physics of the ISM, such as electron temperatures, abundances, ionization fractions, and non-LTE effects (Brown et al. 1978). For the study of AME at 30 GHz, RRLs are useful because they have a direct relationship with the continuum free–free emission (Rohlfs et al. 2013; Alves et al. 2015) in both discrete H ii regions (e.g., Paladini et al. 2003; Bania et al. 2010; Anderson et al. 2011) and the DIG (Luisi et al. 2020; Anderson et al. 2021) from the leakage of Lyα photons from nearby H ii regions (e.g., Zurita et al. 2000).

In addition to measuring continuum emission from the Galactic plane, COMAP can map emission lines within 26 < ν < 34 GHz. In this first analysis we will discuss the five hydrogen RRLs within the COMAP band: H(62)α to H(58)α. The COMAP spectrometer velocity resolution is ∼20 km s⁻¹, just wide enough to resolve these RRLs, which have a typical ${\rm{\Delta }}{V}_{\tfrac{1}{2}}\sim 25$ km s⁻¹. We will focus our attention on the bright region W43, where RRLs have been extensively studied before (e.g., Hoglund & Mezger 1965; Alves et al. 2015; Luisi et al. 2020).

To process the RRLs we select chunks of spectra within ±200 km s⁻¹ around the rest frequencies of the five available RRLs. Then for each observation we fit a first-order polynomial across the spectrum, excluding the velocity range (in the local standard of rest, V_LSR) known to contain W43: 0 < V_LSR < 175 km s⁻¹. To map the spectra we use a simple weighted-averaging map-making method since, unlike with the continuum data, we can assume the data are dominated by white instrumental noise. The calibration, observations, and other data processing are the same as described in Section 2.2.

In Figure 8 we show a map centered on W43 of the integrated RRLs after stacking all five lines; the contours are of the COMAP 26.5 GHz continuum data. The map has units of K km s⁻¹ and a resolution of $4\buildrel{\,\prime}\over{.} 5$. We can clearly see in the center of the map the W43 complex, but we also see the bright H ii region and molecular cloud complex G29.93−0.03, also known as W43-South. Between the W43 and W43-South complexes we can see RRL emission originating in the DIG, which can be seen to be strongly correlated with the continuum emission shown by the contours. Detection of RRLs in the DIG at a similar resolution to that of COMAP has also recently been made at 4–8 GHz (Luisi et al. 2020; Anderson et al. 2021).

Zoom In Zoom Out Reset image size

Figure 9 shows the individual lines measured along the line of sight toward W43 (marked by the dashed yellow circle in Figure 8). We make clear detections of the H(58)α–(62)α lines but we also see evidence of the He transition, which is offset from the main H transition line by −122.166 km s⁻¹. The final panel of Figure 9 shows the average of all five lines. The fitted peak velocity of the region is found to be V = 90.9 km s⁻¹, which agrees with other RRL surveys of W43 (97.5 ± 0.6 km s⁻¹; Alves et al. 2012).

Zoom In Zoom Out Reset image size

It is possible to estimate the electron temperature (T_e ) using the known relationship of the ratio between the integrated line brightness and the underlying continuum emission:

$$\begin{eqnarray}&&\frac{\int {T}_{L}d{V}_{\mathrm{km}\,{{\rm{s}}}^{-1}}}{{T}_{C}}=1.0534\times {10}^{4}\frac{{T}_{e}^{-1}{\nu }_{\mathrm{GHz}}^{1}}{{g}_{\mathrm{ff}}(\nu ,{T}_{e})},\end{eqnarray} \tag{ 18 }$$

where ∫T_L d V is the integrated line emission, T_C is the continuum emission brightness, ν is the frequency, and g_ff(ν, T_e ) is the Gaunt factor given by Equation (11).

In Table 6 we report the measured line velocities, the line-to-continuum ratios, and the derived electron temperatures using Equation (18) for each RRL measured toward W43. We find that the derived electron temperatures are much higher than those previously reported; for example in Alves et al. (2012) the electron temperature of W43 was found to be T_e = 5660 ± 190 K. The reason for this discrepancy is that we do not attempt to correct for contributions to the continuum emission from sources other than free–free emission; hence we are overestimating the continuum emission since there will be some contribution from AME and to a lesser extent from synchrotron emission. An overestimate of the continuum brightness then leads to an overestimate of the electron temperature.

Table 6. RRL Properties of the W43 Region (Figure 8)

Line	V (km s−1)	$\tfrac{{T}_{L}dV}{{T}_{C}}$ (km s−1)	Te (K)
H(62)α	90.9 ± 0.2	8.53 ± 0.15	7810 ± 120
H(61)α	91.2 ± 0.1	8.67 ± 0.10	8060 ± 80
H(60)α	90.8 ± 0.1	9.14 ± 0.09	8080 ± 70
H(59)α	91.1 ± 0.2	10.04 ± 0.09	7840 ± 60
H(58)α	90.4 ± 0.1	9.70 ± 0.07	8460 ± 50

Note. The electron temperature (T_e ) is calculated using Equation (18). V is the fitted peak frequency of the RRL transitions, and $\tfrac{{T}_{L}dV}{{T}_{C}}$ is the line-to-continuum brightness ratio.

Download table as:  ASCIITypeset image

As discussed in earlier sections the two main emission components around 30 GHz will be either free–free emission or AME. By leveraging our knowledge of the electron temperature of W43 from Alves et al. (2012) we can estimate what fraction of the total emission is due to AME. The fractional AME, for a given RRL, can be defined as

$$\begin{eqnarray}&&{\eta }_{\mathrm{AME}}^{\mathrm{RRL}}=\displaystyle \frac{{T}_{\mathrm{AME}}}{{T}_{\mathrm{ff}}+{T}_{\mathrm{AME}}}=1-\displaystyle \frac{{\hat{a}}_{n}}{{a}_{n}}=1-\displaystyle \frac{{T}_{\mathrm{ff}}}{{T}_{\mathrm{ff}}+{T}_{\mathrm{AME}}},\end{eqnarray} \tag{ 19 }$$

where ${\hat{a}}_{n}$ is the line-to-continuum ratio we measure in Table 6 for transition H(n)α, a_n is the expected ratio given by Equation (18) and an assumed T_e = 5660 K, T_AME is the continuum brightness of the AME, and T_ff is the continuum brightness of the free–free emission. We find that the mean fractional AME within W43 at 30 GHz is ${\eta }_{\mathrm{AME}}^{\mathrm{RRL}}=0.35\pm 0.01$, i.e., (35 ± 1)%, which is consistent with previous estimates (e.g., 37% ± 5%; Irfan et al. 2015).

These initial results show that the COMAP survey can readily detect RRLs from the Galaxy. The sensitivity is sufficient to detect several RRLs not only from the brightest H ii regions but also from weaker sources including the DIG from multiple sightlines. Although not a dedicated survey with high velocity resolution, this will be a useful data set for the community.