© The Authors 2025

1. Introduction

Star formation is a fundamental process for mass growth of galaxies, and thus their evolution. While the rate of star formation is closely tied to the amount of molecular gas present on scales ranging from individual cores inside molecular clouds to entire galaxies (e.g., Bigiel et al. 2008; Kennicutt & Evans 2012), (massive) star formation occurs only in the densest regions of molecular clouds (Gao & Solomon 2004; Wu et al. 2005; Lada et al. 2012; Evans et al. 2014). These dense regions within molecular clouds are generally too small to be resolved at extragalactic distances. Therefore, to study molecular gas at different densities, especially in external galaxies, astronomers combine emission lines that span a wide range of critical or effective excitation densities (e.g., Shirley 2015).

The high-dipole moment molecules HCN, its isomer HNC, and HCO⁺ all have high critical densities and relatively bright rotational transitions in the λ = 3 mm window. As a result, these emission lines have often been considered as “dense gas tracers” and have been observed in extragalactic targets for decades (Helfer & Blitz 1993; Aalto et al. 1997, 2002, 2012; Gao & Solomon 2004; Brouillet et al. 2005; Buchbender et al. 2013; Usero et al. 2015; Neumann et al. 2023; Bigiel et al. 2016; Jiménez-Donaire et al. 2019; Krieger et al. 2020; Bešlić et al. 2021; Eibensteiner et al. 2022; Rybak et al. 2022; Imanishi et al. 2023). Many of these surveys reveal systematic variations in line ratios like HCN/CO as a function of the environment (see references above and review by Schinnerer & Leroy 2024), suggesting that the host galaxy sets the initial conditions for star formation. Even further, the efficiency with which stars are formed out of dense gas seems to vary drastically within larger-scale structures, such as larger-scale bars (Bešlić et al. 2021; Eibensteiner et al. 2022; Neumann et al. 2024) or even our own Milky Way’s central molecular zone (CMZ, Longmore et al. 2013).

Unfortunately, these emission lines are also typically ≳20 times fainter than ¹²CO (Usero et al. 2015; Jiménez-Donaire et al. 2019, 2023; Schinnerer & Leroy 2024). As a result, recent observations of those dense gas tracers attempting to link galaxy environment and molecular gas conditions either focus on mapping larger regions of galaxy disks at low resolution (∼500 − 1000 pc; Bigiel et al. 2016; Jiménez-Donaire et al. 2019; Gallagher et al. 2018a; Heyer et al. 2022; Neumann et al. 2023) or focus on higher-resolution (i.e., ∼100 pc) observations of individual regions within galaxies (e.g., galaxy centers, spiral arms; Chen et al. 2017; Bešlić et al. 2021, 2024; Neumann et al. 2024). While the former is insufficient to isolate individual star-forming regions of molecular clouds, the latter does not capture changes across different environments.

Clearly, a major next step in dense gas tracer studies is to observe these tracers at both much higher physical resolution and across larger areas in galaxy disks. This paper presents the IRAM Large Program “Surveying the Whirlpool at Arcseconds with NOEMA” (SWAN), which aims to take this natural next step. SWAN used the NOrthern Extended Millimetre Array (NOEMA) and IRAM 30m single-dish telescope to map the emission from dense gas tracers (HCN(1−0), HNC(1−0), HCO⁺(1−0), N₂H⁺(1−0)) in the 3 mm band across a large 5 × 7 kpc² portion of the prototypical grand-design spiral M51, one of the closest (D ≈ 8.5 Mpc; McQuinn et al. 2016) northern, face-on, star-forming galaxies. Additionally, we observe CO isotopologs C¹⁸O, ¹³CO(1−0), shock-tracing emission lines (HNCO(4−3), HNCO(5−4)), and tracers of photo-dissociation regions (PDR, C₂H(1−0)). We achieve 125 pc resolution, which is sufficient to resolve the population of giant molecular clouds (GMCs) and approaches the size scale of individual massive GMCs or star-forming complexes.

Of particular interest is the first high-resolution, high-sensitivity wide-field extragalactic map of N₂H⁺ from SWAN (Stuber et al. 2023). Galactic studies, which detect a much larger suite of molecular emission lines due to the proximity of their targets, prefer the use of this molecular ion, N₂H⁺, over HCN, HNC, and HCO⁺ to identify regions of dense (n ∼ 10⁵ cm⁻³) molecular gas. Based on Galactic studies, there is an active ongoing discussion in the literature about how the observed intensity of HCN, HNC, and HCO⁺ depends on the gas density distribution, chemical abundances, and other factors (Shirley 2015; Gallagher et al. 2018b; Pety et al. 2017; Kauffmann et al. 2017; Leroy et al. 2017; Heyer et al. 2022; Santa-Maria et al. 2023; Neumann et al. 2023; Tafalla et al. 2023). N₂H⁺, in contrast, not only has a high critical density, but chemical reactions in the molecular phase of the interstellar medium (ISM) ensure that N₂H⁺ is a selective tracer of dense gas. One of the main destruction mechanisms of N₂H⁺ is the reaction with ¹²CO to form HCO⁺ and other molecules. N₂H⁺ therefore exclusively survives in the densest, coldest parts of molecular clouds, where ¹²CO is frozen out onto dust grains (column densities above 10²² cm⁻²; see Pety et al. 2017; Kauffmann et al. 2017; Tafalla et al. 2021). Since N₂H⁺ emission is more than ∼70 times fainter than ¹²CO emission (e.g., Jiménez-Donaire et al. 2023; Stuber et al. 2023), it has been challenging to observe in extragalactic targets. Previous observations of N₂H⁺ in targets beyond the Local Group include a handful of single-target single-dish studies with kiloparsec-scale resolution (e.g., den Brok et al. 2022; Jiménez-Donaire et al. 2023) and a few dedicated higher-resolution studies of bright galaxy centers (e.g., Meier & Turner 2005; Martín et al. 2021; Eibensteiner et al. 2022). In light of the importance of mapping this tracer across a galaxy disk, preliminary maps of N₂H⁺ and HCN from the SWAN survey have already been presented in Stuber et al. (2023, hereafter: S23).

SWAN builds on the high-quality view of the molecular ISM from CO(1−0) mapping by PAWS (Schinnerer et al. 2013; Meidt et al. 2013; Colombo et al. 2014a) as well as overlapping coverage by VLA radio continuum mapping (including free-free emission), JWST, and HST recombination line and infrared mapping (e.g., Kennicutt et al. 2007; Dumas et al. 2011; Querejeta et al. 2019; Kessler et al. 2020). These means that the bulk molecular ISM and recent star formation are uniquely well constrained in this galaxy. The ISM and dynamical environment across the galaxy are also well understood from extensive previous multiwavelength analysis (e.g., Walter et al. 2008; Meidt et al. 2013; Colombo et al. 2014b; den Brok et al. 2022). M51 was targeted by several previous lower-resolution dense gas tracer mapping studies that detected strong environmental variations in, for example, the HCN/CO ratio across the galaxy (at 500−1000 pc resolution; Usero et al. 2015; Bigiel et al. 2016; Heyer et al. 2022). At ∼100 pc resolution, individual regions within M51 were targeted (i.e., Chen et al. 2017; Querejeta et al. 2019). This makes M51 an ideal target to investigate the physical origin of the environmental variations in HCN/CO, SFR/HCN, and similar quantities observed at much larger spatial scales.

This paper presents the full suite of 3 mm lines observed as part of the SWAN IRAM Large Program, and is structured as follows. In Sect. 2, we describe the SWAN observations. The data reduction, imaging, and data quality assessment are presented in Sect. 3. In Sect. 4, we compare the spatial distribution of all the nine 3 mm lines detected in the SWAN dataset, and compare their relative intensity to ¹²CO and previous literature observations. An interpretation of these findings is provided in Sect. 5. We summarize our main conclusions in Sect. 6.

The richness of the SWAN data allows for more detailed studies of, for example, the variations in ¹²CO isotopologs (Galić et al. 2024) and also, in combination with SMA observations in den Brok et al. (2025), a direct comparison of the dense gas tracing lines across different environments (for a first comparison of HCN and N₂H⁺, see Stuber et al. 2023, with an extensive comparison including HNC and HCO⁺ underway; Stuber et al., in prep.) and with cloud properties (e.g., gas mass surface density, line width, virial parameter; Bigiel et al., in prep.). Further dedicated studies of the central molecular outflow are under way (Thorp et al., in prep.; Usero et al., in prep.).

2. Observations

SWAN utilizes observations from the IRAM Large Program LP003 (PIs: E. Schinnerer & F. Bigiel), which mapped emission lines in the ∼3 mm band across the central 5 × 7 kpc² of the nearby galaxy M51 with both the NOEMA and the IRAM 30m single-dish telescope.

2.1. NOEMA observations

Observations with NOEMA were taken between January 2020 and December 2021, with the antenna configuration split between the C (59%; 126 h) and D (41%; 88 h) configurations. In total, we used 17 pointings to map the central ∼5 × 7 kpc² of M51’s disk (Fig. 1). The mosaic uses hexagonal spacing.

Fig. 1. Primary beam of individual mosaic pointings (blue circles) of the NOEMA observations for SWAN overlaid on a map of the integrated intensity of HCN(1−0) emission (left) and 13CO(1−0) emission (right). Contours represent integrated N2H+ emission at 0.5 and 2 K km s−1. The pointings shown have a radius of ∼28″ (HCN) and ∼22.5″ (13CO, our highest frequency detected line). We interpolate the outer edges of the individual mosaic pointings to define the area inside which we measure the integrated flux of each observed emission line (red line). Because the primary beam changes with frequency, so does the area used to calculate the line flux.

For the SWAN observations, the receiver was tuned to a sky frequency of 88.49177 GHz, which corresponds to the frequency of HCN(1−0) redshifted by the systemic velocity of M51 (v_sys ∼ 471.7 km s⁻¹). We covered the ∼15.5 GHz (×2 polarizations) instantaneous bandwidth at the default Polyfix 2 MHz resolution. The setup was selected to cover the main target lines ¹³CO, C¹⁸O(1−0) in the upper side band, as well as HCN, HCO⁺ HNC and N₂H⁺ (1−0) in the lower side band (Table 2). As the configuration of Polyfix permits up to 16 high-resolution (62.5 kHz) windows per 4 GHz, we placed between three and five such windows around the rest frequency of each of our target lines.

We evaluated the quality of the SWAN observations based on the automatic calibration reports obtained with the standard GILDAS/CLIC calibration pipeline. The automatic quality assessment tool was used to filter out poor data (see GILDAS/CLIC manual¹). Out of a total of ∼246 h of observations, ∼214 h were taken under average to excellent observing conditions (i.e., low water vapor, minimum cloud coverage, minimum antenna tracking errors without systematic variations, good phase and amplitude stability). The average water vapor during these 214 h was ∼4 mm.

We used the quasars J1259+516 and J1332+473 as our main phase and amplitude calibrators, substituted by 1418+546 if either calibration target was unavailable. Observations of the calibrators were executed every ∼17 minutes. Absolute flux calibration was performed using IRAM models for MWC349 and LkHa101, providing about 50 independent measurements of the flux of J1259+516 and J1332+473 and quasar 2010+723 over a period of about one year. This allowed us to confirm that variations in the flux of these quasars happen mostly over longer time periods and are relatively smooth. We show the flux of J1259+516, J1332+473, and 2010+723 over time in Fig. 2, inferred by the GILDAS pipeline solution, and after manual adjustments. While the pipeline is deriving the flux gain for each observation separately by ensuring that the flux of a primary flux calibrator is equal to the modeled value, the manual solution uses the temporal stability of the most stable quasar from one day to the next. That means we take advantage of the fact that observations were executed over consecutive days. For sets of observations taken on consecutive days, we identify as new flux reference the quasar whose flux is most stable over this time range. We used its derived flux value from the day with best observing conditions as a reference to solve again for the flux gains of the other observations in each set. Significant changes in the derived flux gains only occurred when the observations of the primary flux calibrator were not taken under good conditions; for example, when the elevation of the primary flux calibrator was low. Since the full mosaic is observed between two phase calibrator observations, the flux calibration has the same effect for all its 17 pointings. The same is true for all lines, as they are observed simultaneously and a global spectral factor is applied per sideband. In short, while the flux calibration impacts the absolute fluxes of the SWAN data, it does not impact the relative strengths between the different lines observed. Based on Fig. 2, we estimate that the absolute flux uncertainty for the NOEMA observations is ∼10%.

Fig. 2. Observed flux (top three rows) and spectral index (bottom three rows) over time for the three used phase and amplitude calibrators of the SWAN NOEMA observations. In addition to the adopted solution (colored points), we show the flux solution determined by the pipeline (gray circles). The temporal variation in the flux of these quasars is relatively smooth. Manual adjustments result in smaller time variations in the spectral index.

2.2. IRAM-30m single-dish observations

Spatially extended emission accounts for a significant fraction of the total flux in galaxies (e.g., Appendix D of Leroy et al. 2021a). Since interferometers are not sensitive to emission from coarser scales, we complement the NOEMA interferometric data with single-dish data from the IRAM-30m telescope to provide the low spatial frequency (or “short-spacing”) information.

The SWAN 30m data partially consists of archival IRAM 30m observations. We used archival HCN(1−0), HNC(1−0) and HCO⁺(1−0) data from the IRAM-30m EMPIRE survey (Jiménez-Donaire et al. 2019), and N₂H⁺(1−0), C¹⁸O(1−0) and ¹³CO(1−0) data from the IRAM-30m CLAWS survey (055−17, PI: K. Sliwa; den Brok et al. 2022). More detailed information about the EMPIRE and CLAWS observations can be found in the corresponding survey papers (Jiménez-Donaire et al. 2019; den Brok et al. 2022). Both surveys cover a field of view (FoV) that is significantly larger than that of the NOEMA observations. Within the NOEMA FoV, the EMPIRE and CLAWS observations represent about 14 hours of 30m observations (see also Table 1), insufficient for our sensitivity requirements and to avoid any degradation when combining 30m and NOEMA data (according to Eq. 19 of memo IRAM-2008-2; Rodríguez-Fernández et al. 2008). We therefore obtained ∼55 hours of new observations with the IRAM-30m between 2020 February and April (project 238-19). M51 was observed with EMIR combined with the Fast Fourier Transform Spectrometers (FTS). We used a frequency bandwidth of 2 × 7.9 GHz, regularly sampled at 195 kHz. We used on-the-fly (OTF) mapping mode with scan legs of lengths 200 and 300″ along the RA and Dec axes, respectively. The distance between two scan legs (i.e., perpendicular to the scanning direction) was 8″. Additionally, we shifted the center of the mapped box in each iteration by multiples of 2″ to get a final grid with a finer step of order 2″.

We typically observed under good weather conditions (median PWV = $2_{-2}^{+4}$ mm, where, hereafter, the subscript and superscript indicate the offsets of the 16^th and 84^th percentiles from the median, respectively) and at high elevation (${69}_{-16°}^{°+10°}$). The typical system temperatures on the $T_{\mathrm{A}}^{\ast }$ scale were ${89}_{-12}^{+10}$ K for lines below 91 GHz, ${79}_{-6}^{+12}$ K for N₂H⁺(1−0), and ${109}_{-16}^{+17}$ K for the CO isotopolog lines and HNCO(5−4).

The absolute flux calibration of the 30m observations was monitored with the bright carbon-rich AGB star IRC+10216, which was observed for a few minutes at the start of most of the observing runs. For each observation, we extracted and integrated ten well-detected lines (signal-to-noise ratio (S/N) ≳50) with frequencies spaced across the observed bandwidth (88.6 − 110.2 GHz). This allowed us to check the relative calibration of the telescope as a function of time and frequency disregarding thermal noise effects. We confirmed that the spectral shape of every line is constant with time up to a varying scaling factor. This suggests that the observed amplitude variations are not caused by pointing errors. In short, since the lines are distributed in different ways across the envelope of IRC+10216, pointing errors would tend to change the shape of the lines with widespread emission. We also found that the correlation between the integrated intensities of pairs of lines was better when considering lines within the same EMIR subband, suggesting that the observed amplitude variations are driven by calibration errors that depend sensitively on frequency. Overall, the root mean square (rms) uncertainty in the integrated intensities is of order ∼5% for most molecular lines studied, and ∼10 − 15% for the four lines in the 93−102 GHz regime. We also detected systematic differences between polarizations of order ∼5 − 10% for all lines. In the end, we can assume a relative flux calibration uncertainty on the order of 5 − 10% for all lines, consistent with typical expectations.

3. Data reduction and imaging

3.1. NOEMA data reduction and imaging

We calibrated the NOEMA observations using the standard GILDAS/CLIC² pipeline. We extracted calibrated uv tables for each velocity resolution (i.e., 10, 5, 2.5, and 1 km s⁻¹) and line before imaging with GILDAS/MAPPING. The exact spectral extent of the resulting cubes depends on the line, as is described in Sect. 2.1. We did not subtract a continuum in the visibilities in order to avoid biasing the very broad line (typically several hundred km s⁻¹) that appears near the galaxy center. The produced cubes thus contain a contribution from continuum emission. However, for the purposes of line analysis, this continuum contribution is only significant in very small regions along the southern nuclear radio jet axis (XNC; Ford et al. 1985). Hereafter, we subtract this contribution in the image plane when needed (see Sect. 3.3). We imaged all the uv tables on the same spatial grid, which we centered at RA = 13:29:52.532, Dec = 47:11:41.982 (J2000). The grid has a pixel size of 0.31″ and a total map size of 768 × 1024 pixels. Cleaning was performed with the Högbom cleaning algorithm with a constant number of clean components. The number of clean components was chosen to obtain residuals that look like noise (i.e., no coherent spatial structures are present any longer). Since the emission of the brighter lines (e.g., ¹³CO) extends across the full FoV we did not use any support (i.e., cleaning mask) during the cleaning. We used 64 000 clean components for the 10 and 5 km s⁻¹ cubes and 32 000 for the 2.5 and 1 km s⁻¹ ones. We used fewer clean components for the higher spectral resolution cubes since the S/N is lower at higher spectral resolution. Finally, we converted the intensity scale from Jansky per beam to Kelvin with the standard GILDAS-MAPPING GO JY2K command.

The final NOEMA dataset yields detections of nine molecular lines between ∼87.1 and 110.3 GHz: C₂H(1−0), HNCO(4−3), HCN(1−0), HCO⁺(1−0), HNC(1−0), N₂H⁺(1−0), C¹⁸O(1−0), HNCO(5−4) and ¹³CO(1−0). We detect the two brightest C₂H(1−0) hyperfine transitions at rest-frequencies of 87.317 and 87.402 GHz, as well as a fainter transition at 87.284 GHz overlapping along the velocity axis of the first transition mentioned. We do not attempt to separate them, but instead generate one data cube that contains all the detected C₂H emission. The spectral axis of the delivered C₂H data cube is centered on 87.3169 GHz (N = 1 − 0, J = 3/2 − 1/2, F = 2 − 1). We list the rest frequencies of our detected lines in Table 2, as well as the typical rms, peak temperature and native resolution of the data cubes with 10, 5, 2.5, and 1 km s⁻¹ spectral resolution.

3.2. 30m data reduction and imaging

The new IRAM 30m data were reduced to a set of common spectral and spatial grids to ensure a homogeneous treatment for all lines. First, the data for each target line was isolated by extracting frequency windows of 350 MHz (∼950 − 1200 km s⁻¹) centered on the rest frequency of each line. For C₂H(1−0), we increased the window width to 400 MHz to ensure that all the hyperfine structure components were included. Using the Ruze formula with the CLASS command MODIFY BEAM_EFF /RUZE, the temperature scale was converted from $T_{\mathrm{A}}^{\star }$ to T_mb. Next, the data were spatially reprojected to match the NOEMA projection center of RA = 13:29:52.532, Dec = 47:11:41.982 (J2000). The Doppler correction was then recomputed for each spectrum to take into account (1) the change of velocity convention from optical during the observations to radio during the analysis, and (2) its variation as a function of position (because the Doppler tracking is computed only at the projection center and kept fixed during each scan to avoid creating standing waves). The velocity scale was updated to ensure that the redshifted frequency of the line in the local standard of rest (LSR) frame corresponds to a systemic velocity of 0 km s⁻¹. The mean rms noise across each data cube was measured using the baseline residuals. We removed spectra for which the rms noise was greater than three times the mean value in the cube (∼3% of all acquired spectra were rejected in this step).

The IRAM-30m observations were imaged using standard GILDAS/CLASS procedures. The data was regridded spectrally to match the NOEMA data, with four different spectral resolutions: 10 km s⁻¹, 5 km s⁻¹, 2.5 km s⁻¹ and 1 km s⁻¹. All were imaged using a Gaussian kernel of FWHM ∼1/3 the 30m HPBW. We produced two sets of 30m data products with different spatial grids. The first grid has a pixel size of 4″ for distribution as a stand-alone product. The second grid is identical to the one used to grid the NOEMA data to ensure a good spatial sampling when merging the single-dish and interferometric datasets. We list the typical peak intensities and noise levels of the SWAN 30m data in Table 2.

We conducted several checks on the 30m data reduction and imaging, which we describe in more detail in the appendices. Specifically, we tested for consistency between the new and archival 30m datasets (Appendix B). Overall, we find that the datasets agree to within 10%. Finally, we analyzed the impact of the 30m error beams onto the flux filtered out by the interferometer in Appendix A. For ¹³CO, the contribution of the error beam is < 10% for the vast majority (i.e., 98%) of sightlines, and the median error beam contribution is 1.5% per pixel inside the NOEMA SWAN FoV. As the main beam efficiency decreases with increasing frequency, we expect the error beam contribution to be less significant for other emission lines in the SWAN survey.

3.3. Combined NOEMA+30m imaging

The 30m data were combined with the NOEMA data in the uv plane using the GILDAS/MAPPING UV_SHORT command (see Pety & Rodríguez-Fernández 2010, for details). For the combination, we used the calibrated but non-continuum-subtracted NOEMA uv tables and the baseline-subtracted single-dish uv tables. This is slightly inconsistent, but there is currently no reliable method to measure the continuum emission at 3 mm with the IRAM 30m. For several lines, this can result in the noise level in the center being on average offset towards slightly positive values (compare this with Fig. 4). For N₂H⁺(1−0) and HNCO(4−3), we apply an order 1 baseline subtraction in the final data cube during post-processing. We confirm that the baseline subtraction does not affect any of the quantitative results for these lines. To verify this, we performed all analyses shown below using both the baseline-corrected and uncorrected data cubes for N₂H⁺. The overall scientific conclusions and observed trends remain unchanged.

Fig. 4. Spectra of all molecular lines for the full disk (left panel), a 3″ sized region in radius in the center (middle panel) and the western edge of the molecular ring (right panel) where N2H+ is particularly bright. The full disk spectra are extracted using the area covered by the perimeter (“hull”) of the mosaic of all pointings (see Fig. 1) for 13CO. Since this hull is frequency dependent, this corresponds to the smallest area out of all lines.

We image the combined data in a similar way as for the NOEMA-only data (Sect. 3.1). The resulting data cubes have the same spatial grid as for the stand-alone NOEMA data (i.e., a pixel size of 0.31″, which is ∼7 − 10 times smaller than the native resolution beam size, on a grid of 768 × 1024 pixels) and cleaned via Högbom-cleaning without cleaning masks until the cleaned flux reaches a stable number. We used twice as many clean components as for the NOEMA-only data (128 000 clean components for the 10 and 5 km s⁻¹ cubes and 64 000 for the 2.5 and 1 km s⁻¹ ones), since the short-spacing data introduces additional complexity. Lastly, we convert the intensity scale from Jansky per beam to Kelvin. We confirmed that the noise is “well behaved” in each NOEMA+30m line cubes. The rms noise level shows little dependence on frequency across the whole bandwidth, if at all, and there is negligible correlation between adjacent channels.

3.4. Flux recovery

In Table 3, we list the fraction of flux recovered by the native-resolution NOEMA observations for different spectral versions of the SWAN data cube for each emission line. The flux recovery was calculated by smoothing and spatially regridding the NOEMA data to match the 30m data (the spectral grid of the NOEMA data is already matched to the 30m data during the imaging), then summing the emission from the central 100 × 100″ across all channels. The ratio of integrated NOEMA flux to 30m flux yields the flux recovery estimate. We limit the FoV during this computation to avoid contribution from the increased noise toward the edges of the mosaic (Fig. 1). For most lines, the NOEMA observations recover ≲50% of the 30m flux at 10 km s⁻¹ spectral resolution. This fraction is consistently lower for all lines when the data products are imaged with narrower channels, falling to a typical value of 20 − 30% flux recovery at a spectral resolution of 1 km s⁻¹. This is probably due to the lower sensitivity of the data at higher spectral resolution, which affects the deconvolution. Our method for calculating flux recovery is not effective for the fainter lines like N₂H⁺, C₂H, and HNCO. Regridding and smoothing the NOEMA-only data of those lines to match the 30m data strongly reduces the line S/N. As a result, we cannot make precise conclusions about these specific lines (uncertainties as large as ≳80% of the flux recovery value). Table 3 only lists the flux recovery for lines with peak temperature to rms ratio of ≳10 at a spectral resolution of 10 km s⁻¹ in the 30m-only data (compare Table 2).

We also studied the relation between synthesized angular resolution and flux recovery. We tapered the data during the data reduction to coarser angular resolutions of 3, 4, and 6″ and recalculated the flux recovery per channel. Figure 3 shows the flux recovery of the native resolution NOEMA ¹³CO(1−0) data compared to the 30m data for each channel as a function of the angular resolution associated with different tapering distance in the uv plane. At fixed spectral resolution, the flux recovery of the NOEMA data improves as the angular resolution increases from 2 to 4″. Above ≳4″, the flux recovery converges and no longer increases with increasing beam size. Our SWAN flux recovery results are consistent with findings for the PAWS survey (Schinnerer et al. 2013), where only a marginal improvement in the flux recovery was reported when the resolution was degraded from 3″to 6″ (Pety et al. 2013). Our tests indicate that all the flux present in the SWAN NOEMA data is deconvolved at scales ≳4″.

Fig. 3. Flux recovery for 13CO(1−0) at 10 km s−1 spectral resolution. The native resolution NOEMA data (cyan line) was convolved and regridded to match the 30m data (black line) in resolution. The flux shown is the summed flux per channel inside the central 100 × 100″. NOEMA data tapered to coarser spatial resolutions of about 3, 4, and 6″ (blue, purple, red lines) have higher recovery rates.

3.5. Signal in the NOEMA+30m data cubes

The NOEMA+30m data were imaged at spectral resolutions of 10, 5, 2, and 1 km s⁻¹. We list the typical rms and peak temperature per channel for each spectral resolution at the native angular resolution for all lines in Table 2.

Figure 4 shows spectra at 10 km s⁻¹ spectral resolution for the combined NOEMA+30m data for all detected molecular lines: ¹³CO(1−0), C¹⁸O(1−0), HNCO(5−4), N₂H⁺(1−0), HCO⁺(1−0), HNC(1−0), HCN(1−0), HNCO(4−3), and C₂H(1−0). We show spectra of the full FoV inside the perimeter (hereafter “hull”) covered by the mosaics (compare this with Fig. 1), as well as inside a 3″ sized region located at the galaxy center and the bright spot at the southwestern edge of the molecular ring (RA 13:29:50.0633, Dec 47:11:25.2040 (J2000)). As expected, most central spectra, especially from HCN, HNC and HCO⁺, are broader than those in the molecular ring, likely due to the complex kinematics due to the active galactic nucleus (AGN)-driven outflow in the galaxy center. The multiple peaks in the C₂H(1−0) spectra are hyperfine transitions.

3.6. Moment map creation

The final SWAN data cubes of combined NOEMA and IRAM-30m data are publicly available³. Integrating all emission within a fixed velocity range at all lines of sight will add noise to the already faint emission. Therefore, a number of masking techniques that select regions and channels within the data cube to integrate emission have been used in the literature and depending on the method used the total recovered flux this will differ (see, e.g., Appendix B in Pety et al. 2013). The advantages and disadvantages of different masking strategies depend on the science objective, and the relative importance of completeness versus avoiding false positives (see, e.g., Leroy et al. 2021a) Utilizing these data cubes at a spectral resolution of 10 km s⁻¹, we test two commonly used methods of moment-map creation.

First, we created moment-maps with the GILDAS/CUBE “Island method” (Einig et al. 2023). The resulting maps are shown in Fig. 5 (and for a better comparison with an alternative method in Fig. C.1) at each line’s native angular resolution. For each line, this method identifies connected structures as follows. We calculated the noise in channels with velocities |v|> 200 km s⁻¹. We then smoothed the data cube with a Gaussian kernel of size B_maj × B_min * PA and calculated the S/N at each pixel based on the smoothed cube and the calculated noise. Next, we identified connected structures above a selected threshold of S/N = 2 in the position-position-velocity (ppv) cube. These structures were applied to the original data cube and emission was integrated over the pixels within those structures. By doing this for each line at its native resolution, we minimized noise that would otherwise be added in the integration process, while still conserving fainter emission from connected structures. Figure 5 (and Fig. C.1) show the moment-0 maps created when selecting structures based on each line individually at their native angular resolution. In addition to moment-0 maps, we created velocity field maps (moment-1), line-width maps (moment-2), peak temperature maps, and associated uncertainty maps, which are available in the public data release.

Fig. 5. Integrated intensity (moment-0) maps of the SWAN dataset (combined NOEMA and IRAM 30m observations) for the J = 1 − 0 transitions of 13CO, C18O, N2H+, HCO+, HNC, HCN, and C2H plus HNCO(J = 4 − 3) and HNCO(J = 5 − 4) at their native angular resolution (∼2.3 − 3.1″). The lines are grouped in different subsets based on their commonly used applications. The maps were created with the GILDAS Island-method (see Appendix C). We show the beam size in the bottom left of all panels as well as a 1 kpc scale bar in the top right panel. The outline of the SWAN FoV is indicated by a dashed ellipse on top of a multicolor HST image (credit: S. Beckwith (STScI) Hubble Heritage Team, (STScI/AURA), ESA, NASA).

In contrast to selecting connected significant emission structures identified in the same data cube that is being used to generate moment maps, another common strategy is to construct a significant emission mask based on a single bright line (e.g., ¹²CO, aka the “prior”) and applying the resulting mask to the data cubes of fainter lines. Since M51 has an unusual center that is mostly devoid of ¹²CO emission (including the ¹³CO and C¹⁸O isotopologs; see Figs. 4 and 5), a ¹²CO-based prior does not accurately capture the emission from all the lines in the SWAN survey field. Bright HCN HNC, and HCO⁺ emission in the center of M51, for example, are evident in Fig. 4. The spectra inside a 3″ sized central aperture (Fig. 4) are comparably broad for all three of these lines, with FWHM ∼ 100 km s⁻¹. We generate another set of moment maps using both ¹²CO and HCN emission to construct a mask. This mask contains regions in which either HCN or ¹²CO is detected. This mask captures the bulk molecular gas distribution best traced by ¹²CO emission outside of the central kiloparsec, as well as the central region, which is best traced by the bright HCN emission (Fig. 5). This is done by using the so-called “PyStructure”⁴ code (den Brok et al. 2022; Neumann et al. 2023) and after convolving the data to a common resolution of 3.05″. High-resolution ¹²CO(1−0) data were taken from the PdBI Arcsecond Whirlpool Survey (PAWS; Schinnerer et al. 2013) and matched to our resolution. Pixels in ppv space were thus selected for integration if either HCN or ¹²CO was detected. The PyStructure code further allows the user to resample the data with hexagonal pixels, which capture the circular beam of the observations well. We hexagonally resampled all data to a matched grid with two hexagons across each beam length. Integrated moment maps were then created by selecting the spectral windows where both ¹²CO and HCN are significantly detected. This data were saved in a numpy table that we refer to as a “PyStructure table” from hereon.

While the former method is best suited when investigating individual lines, the latter is often preferable for the comparison of several lines, since it ensures that the same pixels of the ppv cube are used for integration. However, it tends to increase the noise in the integrated emission maps of fainter lines (e.g., HNCO(5−4), HNCO(4−3) and C₂H(1−0)). This is apparent when we compare Figs. C.1 and C.2, where differences between them are driven by either the difference in methodology, or the difference in resolution (and therefore S/N). As an example, HNCO(5−4) is shown at its native 2.3″ in Fig. C.1, but smoothed to a 3″ resolution in Fig. C.2. As the native resolution of C₂H is ∼3″, differences in the moment maps arise due to the differences in methodology.

Given that both the GILDAS and PyStructure methods are useful for different analysis, we conduct a pixel-by-pixel comparison for ¹³CO in Appendix C. Overall, we find good agreement between both methods, with the GILDAS method recovering more flux than the Pystructure at lower intensities.

The SWAN public data release includes data cubes and the moment maps created with GILDAS and are available on the IRAM Data Management System and the Canadian Advanced Network for Astronomical Research (CANFAR). The PyStructure table is utilized in upcoming SWAN publications analyzing the CO isotopologs (den Brok et al. 2025; Galić et al. 2024) as well as dense gas tracers HCN, HNC, HCO⁺ and N₂H⁺. (Stuber et al., in prep.). The PyStructure table are distributed as a complementary product on the CANFAR.

4. Results: 3 mm line emission across M51

We present an overview of the integrated intensity maps (moment-0) of all lines from the SWAN survey in Fig. 5. The resulting maps from the two different methods of moment map generations (see Sect. 3.6) can be compared via Figs. C.1 and C.2). To compare the emission of all detected molecular lines, we utilized the moment maps created based on common priors (¹²CO and HCN, Fig. C.2), created with the PyStructure-code.

In this section, we compare the spatial distribution of the integrated molecular line emission across the full FoV below. Further, we compare emission of molecular lines against each other (Sect. 4.1), and compare line ratios with ¹²CO between the central 1 kpc and the disk (Sect. 4.2). In Sect. 4.4, we compare the SWAN dataset to high- and low-resolution observations from the literature for individual lines.

To first order, the SWAN maps show that 3 mm molecular line emission is similarly distributed across M51’s inner disk. For all tracers, the emission is bright along the spiral arms and in the molecular ring. Roughly half of the SWAN emission lines are also bright in M51’s center, where an AGN with low-inclined radio jet is located. The notable exceptions are ¹³CO and C¹⁸O, which – similar to ¹²CO (PAWS; Schinnerer et al. 2013) – show relatively faint emission in the central region compared to elsewhere in the inner disk.

Within the disk, the emission of all lines is particularly bright along the western side of the molecular ring, at the base of the southern spiral arm. Some lines, including N₂H⁺(1−0), C¹⁸O(1−0), ¹³CO(1−0) and HNCO(4−3) are brightest in this particular region (RA: 13:29:50.06332, Dec: 47:11:25.20404 (J2000)). Spectra of all lines inside a 3″-aperture centered on this region and centered on the galaxy center (Fig. 4) reveal that most lines reach higher intensities in this region compared to the center. The N₂H⁺ emission in this region was previously studied by S23, who reported unusually high N₂H⁺-to-HCN and N₂H⁺-to-¹²CO ratios compared to elsewhere in M51’s inner disk. Other lines in the SWAN dataset, such as the shock-tracer HNCO, are also bright in this region. This region will be investigated in more detail, using the full suite of SWAN emission lines, in a forthcoming paper (Stuber et al., in prep.).

4.1. Comparison of line intensities

In Fig. 6, we present a pixel-by-pixel comparison of the integrated intensity for all emission line pairs in the SWAN dataset. We highlight data points inside the central, inclination-corrected 1 kpc (in diameter) region, and overlay a linear relation through the average line ratio to aid visual inspection. As is shown in S23, an aperture of 1 kpc captures most of emission that is likely affected by the AGN, and is in agreement with the spatial extent of optical AGN-typical line ratios, X-ray and radio emission (Blanc et al. 2009). We note that within this area, a nuclear bar coexists. This central region contains ∼200 hexagonal pixels. For each panel, we calculate the mean logarithmic line ratio (b = mean(log₁₀(y/x)), with x and y referring to the values on the x and y axes) for pixels where both of the lines involved in the line ratio are significantly detected (> 3σ). All values are listed in Table D.1. Additionally, we list b_cen and b_disk, which refer to the same calculations performed inside and outside the central 1 kpc, respectively. We add the significance (see Appendix D) of the difference of b_cen and b_disk in Table D.1. Only a few pixels in the SWAN FoV show significant HNCO(5−4) detections, so the regression results including this line are highly uncertain. The few detections of this line are located in the center of the galaxy, as well as the southwestern edge of the molecular ring, where all lines are remarkably bright.

Fig. 6. Logarithmic integrated line emission in K km s−1 of all lines compared on a pixel-by-pixel scale. We show the 2D distribution of emission from both significant detections (colored points, > 3σ) and non-detections (gray points, < 3σ). The color scale of both detected and undetected points indicates the point density and is for visual purposes only. We mark pixels inside the central 1 kpc (in diameter, ∼8″) in cyan. The dashed gray line corresponds to a power-law with a slope of unity and offset b being the average line ratio calculated from pixels with significant detections, including the central pixels. We define b = mean(log10(y/x)), with x and y referring to the values on the x and y axes. We further show bcen (dashed blue line) and bdisk (dashed green line) which were calculated using only pixels inside and outside the central 1 kpc, respectively. bdisk and b often overlap. Uncertainties were calculated following Gaussian error propagation and are listed together with all values of b in Table D.1.

We find the following:

• (a)

  Most emission line pairs exhibit a roughly linear (slope of 1) correlation, when pixels in the central 1 kpc are excluded. This is particularly the case for emission lines that are associated with denser molecular gas, such as HCN, HNC, and HCO⁺, which visually follow a linear correlation over 1−2 orders of magnitude. The isotopologs C¹⁸O and ¹³CO are both also visually well correlated with HCN, HNC, and HCO⁺.

• (b)

  In most panels, pixels in the central region show a clear offset relative to pixels elsewhere in the disk, indicating that the line emission is driven by different mechanisms in the disk compared to the center. HCN, HNC, HCO⁺, N₂H⁺, HNCO, and C₂H emission in the center are significantly enhanced compared to both C¹⁸O and ¹³CO emission (compare Table D.1). HCN emission in the center is clearly enhanced compared to all other lines. The enhancement is strongest when compared to ¹³CO, and weakest relative to its isomer, HNC. Line combinations that do not vary significantly between the center and disk regions are C¹⁸O and ¹³CO, N₂H⁺ and HCO⁺, and combinations including the fainter HNCO and C₂H lines. N₂H⁺(1−0) is enhanced in the center compared to C¹⁸O and ¹³CO, but it is fainter compared to HCN, HNC, and HCO⁺.

• (c)

  Correlations in the disk that visually appear to deviate from a linear trend include the trends between N₂H⁺ and most lines except HNCO(4−3), and between C₂H and most lines except ¹³CO and potentially C¹⁸O. We note that these visual trends are mostly driven by the brightest pixels of each line and could be biased by the differences in S/N. As an example, in S23 we measured a super-linear trend between N₂H⁺ and HCN emission (m = 1.2), which was mainly driven by the brightest pixels in N₂H⁺ emission.

Our data show that while all molecular lines are bright in similar regions in the disk (Fig. 5), there are significant variations in the line ratios, both on 125 pc scales and across larger ∼kiloparsec-scale environments such as the center. Moreover, the not exactly linear correlation between several molecular emission lines suggests that these cloud-scale observations are sufficient to detect some variations in the excitation, chemical abundance, and opacity in the molecular gas across M51’s disk. Overall, the variations in line emission ratios between the central 1 kpc, which potentially arise due to the AGN, and the disk are larger than the variations observed elsewhere in the disk.

4.2. Global line ratios with ¹²CO

¹²CO is often used to study the bulk molecular gas distribution in galaxies (i.e., Helfer et al. 2003; Bolatto et al. 2013; Leroy et al. 2021b), because it is relatively abundant and its rotational transitions easily excited, producing bright millimeter-wavelength emission under typical ISM conditions. When studying other usually fainter molecular lines, it is thus often of interest to measure their intensity relative to ¹²CO. In Fig. 7, we show the logarithmic distribution of the emission in the SWAN moment-0 maps divided by the integrated intensity ¹²CO emission (from Schinnerer et al. 2013, see Section 3.6). The distribution of the integrated intensity ratios are shown separately for the full disk and the central 1 kpc and central 0.4 kpc. Here, full disk refers to the area inside the SWAN FoV, where we avoid the increased noise towards the edges of the mosaic (compare this with Fig. 1). We only consider pixels where the respective line and CO are significantly detected (> 3σ). Table 4 provides the average integrated intensity line ratios in the full disk, central 1 kpc and disk excluding the center, as well as their scatter. To assess the effect of the 3-sigma masking on the total flux, we also quote the fraction of masked-to-unmasked flux. Even within the environments, we note large variations in the ¹²CO line ratios (Fig. 7). We estimate how much noise is contributing to this scatter in Sect. 4.3.

Fig. 7. Histograms of line ratios with 12CO for pixels where both the line emission and 12CO emission are significantly detected (> 3σ) inside the full FoV (black) and the central 1 kpc (blue) and central 0.4 kpc (in diameter, red). We mark averages (log10(mean(line/CO))) for the full FoV and central apertures (dashed gray, blue, and red line, respectively). The histograms are normalized to have an integrated area of unity.

As before, the central 1 kpc distribution is clearly offset from the full disk distribution for most lines except for the ¹²CO line ratios with the CO isotopologs. This offset is even stronger for the HCN-to-CO, HCO⁺-to-CO, HNC-to-CO and C₂H-to-CO ratio in the central 0.4 kpc (diameter). This is caused by ¹²CO emission being nearly absent in the very center of the galaxy, presumably due to photodissociation, mechanical evacuation, or radiative transfer effects (Querejeta et al. 2016). The similar behavior of the ¹³CO- and C¹⁸O-to-¹²CO (1−0) ratios is consistent with results from SMA observations (SMA-PAWS; den Brok et al. 2025), which cover the (2−1) transitions of these CO isotopologs. The similar radial trends observed in the ¹³CO-to-¹²CO (1−0) and (2−1) ratios suggest that these transitions are influenced by similar excitation mechanisms. Still, by integrating the SWAN and SMA-PAWS CO isotopolog line observations, the non-LTE modeling analysis in den Brok et al. (2025) indicates opacity variations at cloud scales within the disk. To assess possible effects of increased ¹²CO opacity, we provide histograms of line ratios with ¹³CO in Appendix E. Qualitatively, the same increases for the dense gas tracer to CO line ratios in the central enhancements can be seen.

We discuss this in more detail in Sect. 5. Since both C¹⁸O and ¹³CO show a similar lack of emission in the center as ¹²CO, their distributions of center and disk agree well.

On average, HCN emission is ∼17 times fainter than ¹²CO, but this factor varies strongly between ∼6 in the galaxy center, and ∼20 in the disk. This emphasizes the importance of mapping emission not only in individual environments but across a larger set of environments such as in our SWAN FoV. The faintest lines detected in our data, N₂H⁺(1−0) and HNCO(4−3), are roughly a factor of ∼70 times fainter than ¹²CO in the disk.

4.3. Effect of noise on measured line ratios with ¹²CO

In addition to the variations in ¹²CO line ratios between center and disk, we observe a large scatter in the histograms presented in Fig. 7, reflected by the large percentiles in Table 5. Our goal is to determine whether the scatter in our dataset can be attributed solely to noise, or if it might be linked to physical properties. The 3-sigma clipping threshold applied during the analysis can introduce a bias in the measurement of line ratios and their scatter.

To explore this, we built a simple toy model for which we assume that the line-to-CO ratio remains constant, and is represented by the median line ratio, which we provide in Table 5. This median line ratio was calculated for pixels where both the line and ¹²CO are significantly detected (> 3σ), similar to the mean values in Table 4. In addition, we provide the median absolute deviation, s_measured, which we use as an indication of the scatter. For each pixel, we predicted the expected line intensity based on the ¹²CO intensity and the constant line ratio. We added a Gaussian noise distribution that is based on each pixel’s noise estimate on both the predicted line intensity and the ¹²CO intensity. Next, we calculated the scatter of the line ratio of these two modified intensities. This procedure was repeated 100 times and the median scatter, s_calculated, is quoted in Table 5. For all line ratios, the calculated scatter is several times smaller than the measured one. This implies that the measured scatter cannot be fully explained by noise and physical mechanisms might be contributing to the scatter.

We estimated the physical contribution by assuming a linear dependency of the line ratio and a statistical and physical noise contribution. This leads to $s_{\mathrm{physical}}=\sqrt{s_{\mathrm{measured}}^2-s_{\mathrm{calculated}}^2}$. We list s_physical for all lines in Table 5. We find that noise uncertainties make a minor contribution to the scatter measured for all line ratios, meaning that the variations in line ratios that we measure at 125 pc resolution are mostly driven by physical mechanisms.

The simple assumption of a constant line ratio is not correct for all lines. As is shown by S23, N₂H⁺ depends superlinearly on ¹²CO (power of ∼1.1) and HCN depends sublinearly on ¹²CO (power of ∼0.5). Since most of these trends are driven by a small number of brighter pixels (i.e., surrounding the AGN), their contribution to the scatter is small. To first order, we consider this a reasonable assumption. We conclude that the variations observed are the effect of physical properties changing and not due to variations associated with the noise of the images.

4.4. Comparison of SWAN data with other surveys

Here, we compare the SWAN results with previous observations to (a) test for consistency with high-resolution (∼3″) observations of HCN(1−0) in three individual pointings in M51 (Querejeta et al. 2019), and (b) put the data into context with lower-resolution (∼30″), but larger-FoV observations of HCN, HCO⁺, and HNC from the EMPIRE survey (Jiménez-Donaire et al. 2019) and ¹³CO, C¹⁸O, and N₂H⁺ from the CLAWS survey (den Brok et al. 2022). While the PyStructure table is well suited for the line-by-line comparison done before, the GILDAS moment-maps and cubes are best suited for such a comparison with literature works (compare Sect. 3.6).

4.4.1. Comparison with high-resolution HCN observations

HCN has previously been observed in M51 at a similar resolution (3″) with the IRAM 30m and PdBI interferometer by Querejeta et al. (2019) (Q19) in three pointings in the disk. This survey mapped the HCN(1−0) flux inside 35″-sized pointings centered on M51’s center (RA = 13:29:52.708, Dec = +47:11:42.81 (J2000)), the northern spiral arm (RA = 13:29:50.824, Dec = +47:12:38.83(J2000)) and the southern spiral arm (RA = 13:29:51.537, Dec = +47:11:01.48(J2000)). We convolved the data to a spatial (3.04″) and spectral (10 km s⁻¹) resolution matched to our SWAN data. The dataset from Q19 was integrated via the same GILDAS island method that we used for SWAN (Sect. 3.6). Figure 8 indicates the HCN observations by Q19 as contours overlaid on our SWAN HCN(1−0) map, as well as average spectra from both datasets corresponding to the 35″ apertures.

Fig. 8. Comparison with HCN observations from Querejeta et al. (2019) (Q19). Left panel: SWAN HCN moment-0 map with contours (∼5, 50, 200σ with σ the average noise) of Q19 HCN maps. Both maps are at 3″ resolution, spatially and spectrally regridded to the same grid, and the maps were created via the same GILDAS island method (using just the HCN line). Circles depict radii of 35″ centered according to Q19. Right panels: Average spectra of SWAN 3″ data at 10 km s−1 resolution as well as the matched Q19 data. Spectra are the average flux inside of the circular area shown in the left panel.

We find good agreement between both the spatial and the spectral distribution of the HCN emission in the Q19 and SWAN datasets. The total emission integrated over the matched spectra in the northern region of Q19 represents ∼119% of the total SWAN emission integrated over the spectra inside the same region. This is likely due to the northern region of Q19 slightly exceeding the SWAN FoV, and thus capturing a slightly larger area. For the center, this fraction is ∼96% and for the southern pointing ∼91%.

4.4.2. Comparison of HCN, HCO⁺, and HNC emission from EMPIRE and CLAWS

Observations of HCN, HCO⁺, and HNC at both lower and higher resolution have proven to be crucial for studying the conditions of molecular gas (Helfer & Blitz 1993; Aalto et al. 1997; Meier & Turner 2005; Kohno 2005; Bigiel et al. 2016; Jiménez-Donaire et al. 2019; den Brok et al. 2022; Imanishi et al. 2023; Neumann et al. 2023; Nakajima et al. 2023). To showcase the difference in resolution, Fig. 9 depicts ∼kiloparsec observations from EMPIRE for HCN, HCO⁺, and HNC(1−0) emission, as well as from CLAWS for C¹⁸O, ¹³CO, and N₂H⁺(1−0), and SWAN contours on top. While the EMPIRE/CLAWS maps cover the outer parts of M51 better due to their larger FoV, their coarse resolution misses several structures that we can resolve in SWAN. SWAN shows a clear difference in molecular line emission between the spiral arms and interarm regions. Additionally, in contrast to EMPIRE/CLAWS, in SWAN we can differentiate between the molecular ring and the AGN-impacted center, which are regions with very different physical conditions.

Fig. 9. Integrated emission maps (moment-0) from EMPIRE at 33″ for HCN, HCO+, HNC (1−0) (left columns) as well as from CLAWS at 15−30″ resolution for C18O, 13CO and N2H+ (1−0) (right columns) with SWAN 1, 5, 10 and 15 K km s−1 contours at native resolution (∼3″) on top. Beam sizes of EMPIRE/CLAWS and SWAN are shown in the bottom left corner of each panel (white and red circles, respectively).

5. Discussion

The SWAN survey provides a view of CO isotopologs, dense gas tracing molecular emission lines, and PDR and shock-tracing lines at the sensitivity and spatial resolution (125 pc) required to bridge extragalactic and Galactic studies. The 5 × 7 kpc² FoV of SWAN covers the central region, which hosts an AGN, a nuclear bar, and a molecular ring, as well as spiral arms and the interarm region. Although all lines are prone to different excitation conditions, our high-resolution maps show general similarities between their distributions across the FoV (Figs. 5, C.1, and C.2). All lines are detected along the northern to southwestern side of the molecular ring, with the brighter lines extending well along the spiral arms. This work provides a first analysis of this dataset and we report stark differences between the emission of these lines and their line ratios with ¹²CO between the central 1 kpc and disk.

CO emission is commonly used to trace the bulk molecular gas distribution. Hence, line ratios with ¹²CO are used to gauge the abundance of molecules or estimate the average gas density (Leroy et al. 2017; Jiménez-Donaire et al. 2019). We compare our average CO line ratios from the center, disk, and full FoV to literature values from the Milky Way and to high-redshift studies in Figs. 10 and 11. The error bars indicate the change in the average line ratio when applying different masking. When we only require ¹²CO to be detected, instead of selecting pixels where both the line and ¹²CO are significantly detected (Table 4), the average line ratio with ¹²CO decreases for most lines. The distribution of significantly detected pixels is shown for the full FoV as violins for visual comparison. We discuss the variations in line intensities and CO line ratios within M51 and in comparison to other galaxies below.

Fig. 10. Literature comparison of integrated intensities for the J = 1 − 0 transition of (from top to bottom): HCN, HNC, HCO+, and N2H+ emission compared to 12CO. We show the SWAN integrated emission (light-shaded area) for the full FoV (squares), the central 1 kpc (triangle), and the remaining disk (circle). These values are obtained by integrating emission from pixels where both the line emission and 12CO emission is significantly detected (> 3σ). The distribution of pixels in the full FoV is added as violins. The error bars correspond to the difference between this calculation (both 12CO and the line are significantly detected) and when instead using pixels where 12CO is significantly detected. Average literature values (horizontal dashed lines, calculated on a linear scale) are based on values from Milky Way (Pety et al. 2017; Barnes et al. 2020; Jones et al. 2012) and extragalactic sources: M51 studies at lower resolution (Watanabe et al. 2014) and in the outer spiral arm (Chen et al. 2017); different galaxy averages from EMPIRE (Jiménez-Donaire et al. 2019); ∼100 pc studies in M33 (Buchbender et al. 2013), M31 (Brouillet et al. 2005) and NGC6946 (Eibensteiner et al. 2022); ∼10 pc observation in the LMC (Nishimura et al. 2016a) and 80 pc in the dwarf galaxy IC10 (Nishimura et al. 2016b); ∼15 − 19″ in three nearby galaxies (Takano et al. 2019) and upper limits from z ∼ 3 galaxies (Rybak et al. 2022). The values are basically sorted by their physical resolution from a few parsecs in the Milky Way (left) to several kiloparsecs in external galaxies (right).

Fig. 11. Same as Fig. 10, but for the J = 1 − 0 transition of the CO isotopologs C18O, and 13CO, as well as the HNCO(5−4) and HNCO(4−3) lines. The EMPIRE isotopologs are measured by Cormier et al. (2018) instead of the survey paper. Further, we add CLAWS measurements of M51 (den Brok et al. 2022). The measurements are sorted by physical resolution (increasing to ∼kiloparsec scales at the right side of the plot).

5.1. CO isotopologs ¹³CO and C¹⁸O:

The general deficit of the otherwise abundant ¹²CO, C¹⁸O, and ¹³CO emission in the center (Figs. 5 and 7) might suggest the chemical or mechanical destruction or excitation of these molecules via the jet and associated mechanisms. The former is in agreement with findings by Saito et al. (2022) in the outflow of NGC 1068, where ¹²CO isotopologs are faint and potentially destroyed by dissociating photons and electrons (see also Cecil et al. 2002).

In general, all SWAN average line ratios between ¹²CO and ¹³CO or C¹⁸O agree well with estimates for the Milky Way and nearby galaxies from the literature (Fig. 11). Since ¹³CO is significantly brighter than the other molecules (i.e., N₂H⁺) studied here, masking methods affect the obtained ratio less (compare this with F_line, Table 4). Still, a slight decrease in the line ratio with increased physical resolution might be inferred from Fig. 11, but only a few data points are available. Further, we do not see a large difference in line ratios between the center and disk of M51 that is evident for the dense gas tracers. This is in contrast to findings at 100 pc scales in the active galaxy NGC 3627 (Bešlić et al. 2021) where the ¹³CO-to-¹²CO(2-1) ratio is decreased in the center with respect to outer regions such as bar ends. Since NGC 3627 hosts a larger bar than M51 does, the central dynamics might play a critical role.

The C¹⁸O-to-¹²CO ratio increases with increasing physical resolution (Fig. 11, studies roughly sorted by physical resolution), which indicates that masking might affect the measurements. Interestingly, the SWAN measurements are generally higher than measurements from individual regions in both the Milky Way and other galaxies and are comparable to the kiloparsec measurements from CLAWS (den Brok et al. 2022), which cover a larger FoV including the entire molecular-gas-dominated disk of M51.

5.2. Shock tracer HNCO:

The higher rotational transitions of HNCO in the 3 mm range have been suggested as tracers of low-velocity shocks (e.g., Martín et al. 2008; Kelly et al. 2017). HNCO emission thus might suggest the presence of shocks in M51’s center consistent with the location of a low-inclination radio jet and a dense-gas outflow containing both ¹²CO and HCN emission (Querejeta et al. 2019). This is in agreement with findings by Martín et al. (2015), who report enhanced HNCO emission in the cicumnuclear disk surrounding the AGN in NGC 1097 at ∼100 pc resolution, and observations of HNCO arising from two lobes at both sides of the AGN of the Seyfert galaxy NGC 1068 (Takano et al. 2014).

The average HNCO(4−3)-to-¹²CO line ratios of M51’s center and disk are comparable to the high-resolution study in the outer arm of M51 (Chen et al. 2017), as well as the lower-resolution study in M51 (Watanabe et al. 2014) and other galaxy measurements (NGC 253, NGC 1068, and IC 342; Takano et al. 2019). As HNCO(5−4) emission is only detected from very few sight lines, the error bar on its average line ratio with ¹²CO is very large (Fig. 11). Masking pixels without significant detections elevates the HNCO(5−4) average significantly (compare this with the error bar in Fig. 11, which showcases the average when including non-detections). With only 9% of the total ¹²CO flux found in the area where HNCO(5−4) is significantly detected (Table 4), we caution the use of this masked full FoV HNCO(5−4)-to-¹²CO ratio.

5.3. Dense gas tracing lines – HCN, HNC, HCO⁺, N₂H⁺:

While some molecules such as C¹⁸O and ¹³CO (and ¹²CO) are very faint in the center, other lines, such as HCN(1−0), are very bright (Figure 6, Table D.1) and enhanced compared to ¹²CO emission (Fig. 7, Table 4) or ¹³CO emission (Fig. E.1). This enhancement is increased when considering even smaller central apertures, suggesting that the outflow and/or AGN could be the potential cause (Fig. 7).

M51 has long been known to exhibit increased HCN emission in the very center both from kiloparsec-scale studies (Jiménez-Donaire et al. 2019) and those at 30 and 100 pc resolution (Helfer & Blitz 1993; Matsushita et al. 2015; Querejeta et al. 2019). Infrared pumping, weak HCN masing, and an increased HCN abundance or electron excitation in the X-ray-dominated region (XDR) of the AGN are some possibilities suggested throughout the literature (e.g., Blanc et al. 2009; Matsushita et al. 2015; Querejeta et al. 2016; Goldsmith & Kauffmann 2017; Stuber et al. 2023). HCN might potentially partake in the outflow driven by M51’s AGN and radio jet. Similarly, bright HCN emission is colocated at the location of the AGN-driven outflow in the center of the galaxy merger NGC 3256 (Michiyama et al. 2018; Harada et al. 2018), in Mrk 231 (Aalto et al. 2012), in the starburst galaxy NGC 251 (Bešlić et al. 2021), in NGC 1068 (Saito et al. 2022), and in the center of NGC 4321 (Neumann et al. 2024). While both NGC 1068 and M51 host a small weak radio jet, there are differences: no N₂H⁺ emission is detected from the outflow in NGC 1068 (Saito et al. 2022), while bright N₂H⁺ emission is evident in M51’s center and outflow, with increasing average N₂H⁺-to-CO ratios for smaller central apertures (Fig. 7).

Shocks might also be an effective way to destroy or excite ¹²CO molecules to higher states. A potentially young or weak jet might not yet have been able to mechanically remove large quantities of molecular gas. We might be viewing an early stage of molecular gas destruction, where lower density molecular gas such as ¹²CO, which usually covers a larger volume than for example HCN molecules, is destroyed in large quantities, while HCN, which is brighter in denser regions (smaller volume) is not yet affected, or even shock-enhanced (i.e., weak maser or abundance increase). However, dynamical age estimates of the jet or central objects in M51 range from ∼10⁴ − 10⁵ yr (Ford et al. 1985; Matsushita et al. 2007) up to a few Myr (Rampadarath et al. 2018). The destruction of ¹²CO might allow molecules such as N₂H⁺ to form more abundantly as they would otherwise rapidly react with ¹²CO (Bergin & Tafalla 2007). Dissociation of other molecules by the radio jet might provide a large quantity of free electrons in the center of M51, which then leads to an efficient dissociative recombination of N₂H⁺. The exact mechanisms driving the line emission in M51’s center will be discussed in more detail in Thorp et al. (in prep.) and Usero et al. (in prep.).

Similarly to M51, Meier & Turner (2005) find bright N₂H⁺ emission in the center of IC342 at 5″, which they explain by an increased cosmic ray ionization rate or an enhanced N₂ abundance. An increased N₂H⁺ abundance could then promote the formation of HCO⁺ which can form from N₂H⁺ molecules (Harada et al. 2019). This is consistent with the fact that HCO⁺ emission in the center follows a significantly different relation with HCN emission and is more tightly related to N₂H⁺ than HCN (compare Appendix D). This is in agreement with findings by Butterworth et al. (2024) that HCO⁺ and its isotopologs have larger column densities than HCN in the starbursting center of NGC 253 from the ALMA Comprehensive High-resolution Extragalactic Molecular Inventory (ALCHEMI) survey (Martín et al. 2021). Given that neither IC342, nor NGC 253 host an AGN, and M51 does not host a starburst in its center, the similarities between these galaxy centers are puzzling.

Emission from HCN and its isomer HNC are tightly correlated and show only a weak difference in their relation between the center and disk, suggesting that the chemical conditions able to convert one molecule into the other are not changing between the two environments. This is in agreement with findings by Meier & Turner (2005) in IC342. A detailed study of the cloud-scale variations in the HCN-to-HNC ratio will be presented in Stuber et al. (in prep.).

Figure 10 shows a large spread between the average ratios of the dense gas tracers to ¹²CO from the literature and SWAN, especially for HCN, HNC, and HCO⁺. Milky Way studies at high resolution report lower values compared to SWAN, except for the CMZ (Jones et al. 2012), which is consistent with our central averages being increased for all lines. The variation between multiple lines per literature study (i.e., average HCN/CO ratio compared to the HCO⁺/CO line ratio for a single study) is typically smaller than the variation in one line ratio across all studies (e.g., HCN/CO varies strongly from study to study). This might indicate that different masking techniques drastically change the resulting ratios. While some studies apply cloud-finding algorithms to isolate individual clouds, other studies quote full FoV averages. Studies at kiloparsec-resolution such as Watanabe et al. (2014) in the center and southwestern molecular ring in M51 and the EMPIRE survey (Jiménez-Donaire et al. 2019) in general report lower line ratios, and so do high-resolution Milky Way studies. Generally, we find no clear trend with resolution (Fig. 10, studies are basically sorted by resolution).

As can be seen from Fig. 7, the pixel-by-pixel distribution of the line ratios with ¹²CO generally spans over at least one or even two orders of magnitude for most lines. In Sect. 4.3, we estimated how much of the scatter can be attributed to noise and found that there is significant scatter in all line ratios with ¹²CO that cannot be explained by noise only. Several studies report dependencies of ratios between dense gas tracers and ¹²CO on dynamical equilibrium pressure or stellar surface density (Usero et al. 2015; Querejeta et al. 2019; Neumann et al. 2023). A strong radial dependency is found in EMPIRE (Jiménez-Donaire et al. 2019), and their line ratios with ¹²CO are generally enhanced in galaxy centers compared to disks (Fig. 10). Still, the difference between their centers and disks is smaller than our SWAN variations between center and disk, despite EMPIRE covering a larger FoV. HCN, HCO⁺, and HNC emission in the center of NGC 6946 (Eibensteiner et al. 2022) sit between our disk and center measurements, despite NGC 6946 not hosting an AGN. Elevated central line emission is a feature common to galaxies with and without an AGN (Usero et al. 2015; Bigiel et al. 2016; Gallagher et al. 2018a; Jiménez-Donaire et al. 2019; Heyer et al. 2022; Neumann et al. 2024) and attributed to the gas-rich high surface density common to galaxy centers. Studies at a similar resolution to ours in M 33 and M 31 both find significantly lower HCN and HCO⁺, which they consider a result of the sub-solar metallicity in both M 33 and M 31 (Buchbender et al. 2013). The variation in dense gas tracers N₂H⁺, HCO⁺, HCN, and HNC with various physical quantities on cloud scales is investigated in detail by Stuber et al. (in prep.).

Overall, we find significant differences that can arise between the center (where an AGN, outflow, and a nuclear bar are present) and the disk of M51, whereas the variations in the disk are more subtle, but significant (Sect. 4.3). The exact mechanisms affecting the molecular gas in the center and in particular the outflow will be studied in more detail in forthcoming papers (Usero et al., in prep.; Thorp et al., in prep.), we emphasize that all lines except ¹²CO and its isotopologs show enhanced emission in the center. Further, we find large variations in literature ¹²CO line ratios across various resolutions and targets. The least variations are seen for ¹³CO-to-¹²CO, which is consistent across the literature. Both the selected environment (center compared to disk) as well as the masking methods likely influence the line ratios.

6. Summary

We present the first-of-its-kind high-resolution (≲125 pc), high-sensitivity map of 3 mm lines covering an area of ∼5 × 7 kpc² in the inner disk of the Whirlpool galaxy. We detect emission from the CO isotopologs ¹³CO and C¹⁸O(1−0) and the dense gas tracing lines HCN, HNC, HCO⁺, and across the largest FoV to date N₂H⁺(1−0). In addition, we detect HNCO(5−4), (4−3) and hyperfine transitions of C₂H(1−0) in the center and molecular ring of M51. Comparing the emission of those lines to each other and to ¹²CO(1−0) emission from PAWS (Schinnerer et al. 2013) at matched resolution, we find the following:

1. The high-resolution maps show general structural similarities across all molecular lines: bright emission of all molecular lines is detected along the western side of the molecular ring. Emission of the brighter lines is well detected in the southern and northern spiral arm, and all lines except the CO isotopologs are bright in M51’s center.

2. We calculated typical ratios of line brightness with ¹²CO brightness inside the full FoV for pixels where each line is significantly detected. We find the highest value for the ¹³CO-to-¹²CO ratio (0.125), followed by HCN-to-¹²CO (0.059), HCO⁺-to-¹²CO (0.039), C¹⁸O-to-¹²CO (0.03), HNC-to-¹²CO (0.027), C₂H-to-¹²CO (0.019), N₂H⁺-to-¹²CO (0.017), and HNCO-to-¹²CO (5−4: 0.014, 4−3: 0.014).

3. Emission of HCN is significantly enhanced in the central 1 kpc compared to all other detected lines, and emission of ¹²CO isotopologs is significantly reduced in the center compared to non-isotopologs. The only line combinations that do not exhibit a significant offset relation to each other between the center and the disk are the isotopologs ¹³CO and C¹⁸O, the molecular ions N₂H⁺ and HCO⁺, and combinations including the faint HNCO and C₂H lines. ¹²CO line ratios are increased in the central 1 kpc compared to the remaining disk for all lines except C¹⁸O. The largest difference can be seen for the HCN-to-¹²CO line ratio, which is more than a factor of three times larger in the central 1 kpc compared to the disk. HNCO emission might suggest the presence of shocks in the galaxy center, linked to a low-inclined radio jet and a dense-gas outflow. This points to complex conditions, possibly involving increased HCN abundance, infrared pumping, weak HCN masing, or electron excitation in the AGN’s XDR.

4. Line ratios with CO qualitatively compare well to lower-resolution literature studies in other galaxies; that is, we find a similar increase in CO line ratios for HCN, HNC, HCO⁺, and N₂H⁺ in the centers of other galaxies. Still, the overall spread in CO line ratios across Galactic and extragalactic sources is significantly larger than both the differences between center and disk in M51, and the differences between different lines (such as HCN, HNC, HCO⁺, and N₂H⁺). We find that the scatter of SWAN ¹²CO line ratios cannot be explained solely by noise and is likely attributed to local physical mechanisms.

SWAN allows us to study the molecular gas properties at cloud-scale resolution across multiple environments in the iconic Whirlpool galaxy. Dedicated studies focusing on the relationship of ¹²CO isotopologs (den Brok et al. 2025; Galić et al. 2024), the dense-gas tracing lines (Stuber et al., in prep.), and the outflow in the center of M51 (Usero et al., in prep.; Thorp et al., in prep.) will showcase the utility of this rich dataset for investigating physical conditions in the ISM.

Data availability

The data is publicly available at the IRAM data management system, which can be accessed via the following link: https://oms.iram.fr/oms/?dms=showprograms

Acknowledgments

This work has been carried out as part of the PHANGS collaboration and made use of data from the IRAM large program ‘Surveying the Whirlpool galaxy at Arcseconds with NOEMA’ (SWAN). We thank the referee for their valuable and constructive feedback. SKS acknowledges financial support from the German Research Foundation (DFG) via Sino-German research grant SCHI 536/11-1. JdB acknowledges support from the Smithsonian Institution as a Submillimeter Array (SMA) Fellow. CE acknowledges the support of the Jansky Fellow of the National Radio Astronomy Observatory. DL acknowledges the support from the Strategic Priority Research Program of the Chinese Academy of Sciences, grant No. XDB0800401. HAP acknowledges support from the National Science and Technology Council of Taiwan under grant 110-2112-M-032-020-MY3. AU acknowledges support from the Spanish grant PID2022-138560NB-I00, funded by MCIN/AEI/10.13039/501100011033/FEDER, EU.

References

Appendix A: Error beam contribution

Our test show that the 30m error beams only make a small contribution to the flux filtered out by the interferometer. We can describe the response of the 30m telescope to a point source by a set of 2D Gaussians representing the main beam and a set of error beams. As a result, depending on the morphology of the science target, the measured signal might be boosted by emission from beyond the region the telescope is pointing at as the wider side lobes pick up signals from other parts of the sky. For instance, in the case of M51, den Brok et al. (2022) demonstrated that up to 20% of the ¹²CO(2–1) emission in the interarm regions (which sit between two brighter spiral arms) can be accounted for by contributions from the error beams. Kramer et al. (2013) provide an approximation of the 30m telescope beam pattern at different frequencies, which we use as a good first-order approximation.

Given a model for the telescope’s beam pattern, several deconvolution schemes exist that provide an estimate for the error beam free signal (for example Westerhout et al. 1973; Bensch et al. 1997; Lundgren et al. 2004; Pety et al. 2013; Leroy et al. 2015). To quantify the relevance in the case of the set of lines observed by SWAN, we perform the procedure described in den Brok et al. (2022) for ¹³CO(1–0), where we expect the effect to be the largest (as the main beam efficiency decreases with increasing frequency).

A.1. Mathematical framework

We provide a brief overview of the mathematical framework and the method we use to deconvolve the 30m data to estimate the error beam contribution. For details on the calculations, we refer the reader to den Brok et al. (2022). The key parameters are the observed main beam temperature, T_mb (which includes also contributions from the error beam), and the error beam free main beam temperature, ${\widehat{T}}_{\mathrm{mb}}$. They are related via a convolution kernel K as follows:

$$\begin{array}{c}\hfill T_{\mathrm{mb}}=({\delta }^{2\mathrm{D}}+K)\otimes {\widehat{T}}_{\mathrm{mb}},\end{array}$$(A.1)

where “⊗" represents the 2D convolution operation and δ^2D the Dirac 2D distribution. The kernel K contains the sum of all error beam contributions after deconvolution with the main beam.

The particular deconvolution, to obtain the error beam free signal, can be expressed using the Fourier Transform operation, ℱ:

$$\begin{array}{c}\hfill {\widehat{T}}_{\mathrm{mb}}={\mathcal{F}}^{-1}\left(\frac{\mathcal{F}(T_{\mathrm{mb}})}{1+\mathcal{F}(K)}\right).\end{array}$$(A.2)

We perform this calculation in Python with the unsupervised Wiener-Hunt deconvolution. This function estimates the hyperparameters automatically (see den Brok et al. 2022).

A.2. Error beam contributions for ¹³CO(1–0)

We present an overview of the resulting deconvolution of the entire ¹³CO(1–0) SWAN 30m data-only cube in Fig. A.1. We compute the contribution, which is the difference between the measured and the error beam free spectrum, for each line of sight where S/N> 3 for ¹³CO(1–0). The error beam contribution is negligible in the center and along the spiral arms, being 2–3%. The value is elevated to around 5–10% in the interarm region. This is expected because when the telescope points to the interarm regions, part of the spiral arm will be covered by the side lobes, boosting the signal via the error beams. In Fig. A.1, we also illustrate the effect for spectra extracted within six different apertures, from which four are in the interarm and two in the spiral arm region. The error beam contribution is presented by the brown spectrum in each panel. The percentage contribution is computed within the mask that contains the signal (illustrated in blue). With 6% and 7%, pointings 3 and 4 show the largest contributions in this selection of pointings.

Fig. A.1. Quantifying errorbeam contributions to the observed 13CO(1–0) emission. (Left) The map presents the percentage errorbeam contribution to individual sightlines for the 30m only data. We only consider sightlines where the integrated 13CO(1–0) brightness temperature is detected at ≳3σ. Contours indicate the S/N at 5,10,20, and 50. The red contour illustrates the SWAN FoV. Overall, the error beam contribution is marginal, with elevated values up to 10% within the interarm region (and < 5% within the SWAN FoV). (Right) We extract 13CO(1–0) spectra in six individual apertures. Apertures 1–4 are in the interarm region and apertures 5–6 in the spiral arms of M51. Each panel shows the full spectrum which includes the contribution from the error beam in blue. The brown spectrum represents the error beam contribution and was calculated by subtracting the error beam free spectrum from the observed spectrum. The percentage contribution is computed over the spectral range where we detect emission (indicated by the blue-shaded region) and is listed in each panel.

The error beam analysis is subject to uncertainties due to the difficulty of characterizing the variations in the beam pattern with time. However, these calculations provide a reasonable upper limit for the order of magnitude of the error beam contribution. The contribution remains < 10% for the vast majority (i.e., 98%) of sightlines. Within the NOEMA SWAN FoV, the median error beam contribution is 1.5% per pixel. As power within the side lobes decreases with decreasing frequency, the effect will be even less significant for the other lines, such as HCN and N₂H⁺.

Appendix B: Consistency tests of the archival and new SWAN 30m datasets

For this comparison, both the SWAN and EMPIRE 30m data were processed in the same way using the EMPIRE pipeline (see Jiménez-Donaire et al. 2019). The raw spectra were first calibrated to antenna temperatures scale using the nearest chopper-wheel calibration scan. After this, each observed line was extracted using the CLASS package. For each individual line of sight for which a spectrum is extracted, we subtracted a zeroth-order baseline, regridded the spectrum to a 4 km s⁻¹ channel width, and wrote them out as FITS tables. The spectra were then processed with an IDL procedure that allows us to flag and discard pathological data. We then fit a baseline excluding a velocity window determined from the much brighter mean CO(1-0) emission, to avoid including channels that potentially contain signal. In addition, two more windows adjacent to the central one were included to fit a second-order polynomial baseline, which was then subtracted from the entire spectrum. The final spectra were then sorted according to their measured rms on the line-free windows, relative to the expected value from the radiometer equation, and the highest 10% were rejected (see Jiménez-Donaire et al. 2019, for a detailed description of each step in the pipeline). Finally, all data corresponding to each spectral line were gridded into a cube. We then convolved each data cube to a common working resolution of 33″, using a Gaussian kernel.

Figure B.1 shows a comparison between the main dense gas products obtained with SWAN and EMPIRE 30m observations, processed using the EMPIRE pipeline described above. These include the main lines HCN(1–0) (left panel), HCO⁺(1–0) (middle panel), and HNC(1–0) (right panel). The red line indicates a 1-1 correlation between the EMPIRE and the SWAN datasets. As can be seen from the figure, both datasets agree well for the three different lines. We quantify this by calculating the relative sum of all differences between both datasets. We find that the HCN(1–0), HCO⁺(1–0), and HNC(1–0) measurements agree between EMPIRE and SWAN within 1%, 7%, and 12%, respectively.

Fig. B.1. Pixel-by-pixel comparison of the HCN (1-0), HCO+ (1-0), and HNC (1-0) data cubes obtained with SWAN and EMPIRE IRAM-30m observations. The red line indicates a 1-1 correlation between the EMPIRE and the SWAN datasets. Both single-dish datasets are processed with the EMPIRE pipeline for comparison and overall agree with each other.

Appendix C: Comparison of moment map integration techniques

Given both the GILDAS and PyStructure methods are useful for different analysis, we aim to confirm a general agreement between both methods. To do so, we re-grid the ¹³CO moment-0 map produced with GILDAS and has rectangular pixels, to match the hexagonal pixels of the moment-0 map produced with the PyStructure code. This is done using parts of the PyStructure code. Figure C.3 shows the emission in each hexagonal pixel from the re-gridded GILDAS-moment-0 map compared to that from the moment-0 map that was inherently produced with the PyStructure code. The moment maps generated by the GILDAS method recovers about 25% more flux than the Pystructure method (Fig. C.3). The median flux difference between both maps is 0.9 K km s⁻¹ with 16^th and 84^th percentiles of 0.3 and 2.1 K km s⁻¹. For pixels at high intensities, both methods agree well. The PyStructure method is more conservative and misses broader linewings. As the PyStructure table used for the scientific analysis in this paper is based on ¹²CO and HCN as priors (see above), we do not expect to miss any significant emission.

Fig. C.1. Integrated intensity maps (moment-0) of the SWAN dataset (combined NOEMA and IRAM 30m observations) for the J = 1–0 transitions of 13CO, C18O, N2H+, HCO+, HNC, HCN, C2H(1–0 and HNCO(J = 4–3) plus HNCO(J = 5–4) at their native angular resolution (∼2.3 − 3.1″). The maps were created with the GILDAS Island-method. We show the beam size in the bottom of all panels as well as a 1 kpc scale bar in the top left panel.

Fig. C.2. Integrated intensity maps (moment-0) of the SWAN dataset (combined NOEMA and IRAM 30m observations) for all detected lines at a common resolution of 3″ (125 pc). The data is binned with hexagonal spacing with the PyStructure code (den Brok et al. 2022; Neumann et al. 2023). Spectral windows for the creation of the moment maps are selected based on significant detections of 12CO emission from PAWS (Schinnerer et al. 2013) and HCN(1-0) emission. We show the beam size, a 1 kpc scale bar and mark the central 1 kpc circular area (green points) in the top left panel. The intensity scale is the same as in Fig. C.1.

Fig. C.3. Pixel-by-pixel comparison of the obtained integrated line emission using two different methods for the 10 km s−1 resolution 13CO data cube at native angular resolution. We show pixels located inside (black circles) and outside (gray circles) the hull of the mosaics (compare this with Fig. 1). We show the 1:1 relation (dashed orange line). We mark the average 5σ noise level for both lines (dashed gray line). We note that due to logarithmic spacing, data points containing noise with negative fluxes are not visible. Although this applies to most data points in the interarm region near the edges of our FoV, we emphasize that this comparison is intended to assess how both methods handle regions with significant detected emission, as these areas are typically the focus of scientific analysis. Regions with significantly detected emission is found mostly in the center, the molecular ring, and on the spiral arms.

Appendix D: Average logarithmic line ratios in comparison

We list mean logarithmic line ratios (b) from Fig. 6 (Sect. 4.1) for all detected lines in Table D.1. b was calculated between two lines (line x and line y) in pixels where both lines are significantly detected (> 3σ). b is defined as b = mean(log₁₀(y/x)), with x and y referring to the values on the x and y axes from line x and line y, respectively. Additionally, we provide b_cen and b_disk, which is the same calculation performed on pixels inside and outside the central 1 kpc, respectively. To estimate how strong the central effects on the line ratios are, we calculate σ_{cen − disk}, which is the significance of the difference of b_cen and b_disk, defined as ${\sigma }_{\mathrm{cen}-\mathrm{disk}}=\frac{b_{\mathrm{cen}}-b_{\mathrm{disk}}}{\sqrt{\mathrm{\Delta }{b}_{\mathrm{cen}}^2+\mathrm{\Delta }{b}_{\mathrm{disk}}^2}}$. There are only very few pixels with significant HNCO(5-4) detections, which might bias the calculation of σ_{cen − disk}.

With this, we find the following: The most extreme difference (σ_{cen − disk} > 50) in mean logarithmic line ratios between center and disk can be found for HCN and ¹³CO, as well as for HNC and ¹³CO. This is followed by line ratios between HCN and HCO⁺, HCN and C¹⁸O, HNC and C¹⁸O, HCO⁺ and C¹⁸O, C₂H and C¹⁸O (30 < σ_{cen − disk} < 50). The most extreme differences are therefore seen between the ¹²CO isotopologs and most other lines. This is consistent with the visual lack of ¹³CO and C¹⁸O emission compared to all other detected lines in the galaxy center seen in Fig. C.2.

Most other line combinations have significantly different line ratios in the center compared to the disk, but with a lower value of σ_{cen − disk} < 30. For HCN, the strongest offset relation between center and disk is with C¹⁸O emission, the weakest with isomer HNC and the faint HNCO(5-4) line.

Line combinations that do not exhibit a clear offset relation between center and disk are N₂H⁺ and HCO⁺, C¹⁸O and ¹³CO, and combinations including the HNCO and C₂H lines. All line combinations with ¹³CO show a significant offset relation between center and disk, with the only exception being the other ¹²CO isotopolog, C¹⁸O.

Appendix E: ¹³CO line ratios

We show histograms of line ratios with ¹³CO emission in the full FoV, central 1 and 0.4 kpc (diameter) in Fig. E.1. In agreement with the ¹²CO line ratios (Fig. 7), we see the same qualitative behavior: Line ratios in the center are increased for all lines except the isotopolog C¹⁸O. This increase is even stronger for the smaller (0.4 kpc) aperture for HCN, HNC, HCO⁺, N₂H⁺, and C₂H. We note that the average distribution of the HNCO(5-4)/¹³CO line does not vary much between 1 and 0.4 kpc. As the HNCO(5-4) emission is only detected in few pixels in the very galaxy center, both histograms might depict the same information.

Fig. E.1. Same as Fig. 7 but for line ratios with 13CO(1-0).

All Tables

All Figures

Fig. 1. Primary beam of individual mosaic pointings (blue circles) of the NOEMA observations for SWAN overlaid on a map of the integrated intensity of HCN(1−0) emission (left) and 13CO(1−0) emission (right). Contours represent integrated N2H+ emission at 0.5 and 2 K km s−1. The pointings shown have a radius of ∼28″ (HCN) and ∼22.5″ (13CO, our highest frequency detected line). We interpolate the outer edges of the individual mosaic pointings to define the area inside which we measure the integrated flux of each observed emission line (red line). Because the primary beam changes with frequency, so does the area used to calculate the line flux.

In the text

Fig. 2. Observed flux (top three rows) and spectral index (bottom three rows) over time for the three used phase and amplitude calibrators of the SWAN NOEMA observations. In addition to the adopted solution (colored points), we show the flux solution determined by the pipeline (gray circles). The temporal variation in the flux of these quasars is relatively smooth. Manual adjustments result in smaller time variations in the spectral index.

In the text

Fig. 4. Spectra of all molecular lines for the full disk (left panel), a 3″ sized region in radius in the center (middle panel) and the western edge of the molecular ring (right panel) where N2H+ is particularly bright. The full disk spectra are extracted using the area covered by the perimeter (“hull”) of the mosaic of all pointings (see Fig. 1) for 13CO. Since this hull is frequency dependent, this corresponds to the smallest area out of all lines.

In the text

	Fig. 3. Flux recovery for 13CO(1−0) at 10 km s−1 spectral resolution. The native resolution NOEMA data (cyan line) was convolved and regridded to match the 30m data (black line) in resolution. The flux shown is the summed flux per channel inside the central 100 × 100″. NOEMA data tapered to coarser spatial resolutions of about 3, 4, and 6″ (blue, purple, red lines) have higher recovery rates.
In the text

Fig. 5. Integrated intensity (moment-0) maps of the SWAN dataset (combined NOEMA and IRAM 30m observations) for the J = 1 − 0 transitions of 13CO, C18O, N2H+, HCO+, HNC, HCN, and C2H plus HNCO(J = 4 − 3) and HNCO(J = 5 − 4) at their native angular resolution (∼2.3 − 3.1″). The lines are grouped in different subsets based on their commonly used applications. The maps were created with the GILDAS Island-method (see Appendix C). We show the beam size in the bottom left of all panels as well as a 1 kpc scale bar in the top right panel. The outline of the SWAN FoV is indicated by a dashed ellipse on top of a multicolor HST image (credit: S. Beckwith (STScI) Hubble Heritage Team, (STScI/AURA), ESA, NASA).

In the text

Fig. 6. Logarithmic integrated line emission in K km s−1 of all lines compared on a pixel-by-pixel scale. We show the 2D distribution of emission from both significant detections (colored points, > 3σ) and non-detections (gray points, < 3σ). The color scale of both detected and undetected points indicates the point density and is for visual purposes only. We mark pixels inside the central 1 kpc (in diameter, ∼8″) in cyan. The dashed gray line corresponds to a power-law with a slope of unity and offset b being the average line ratio calculated from pixels with significant detections, including the central pixels. We define b = mean(log10(y/x)), with x and y referring to the values on the x and y axes. We further show bcen (dashed blue line) and bdisk (dashed green line) which were calculated using only pixels inside and outside the central 1 kpc, respectively. bdisk and b often overlap. Uncertainties were calculated following Gaussian error propagation and are listed together with all values of b in Table D.1.

In the text

Fig. 7. Histograms of line ratios with 12CO for pixels where both the line emission and 12CO emission are significantly detected (> 3σ) inside the full FoV (black) and the central 1 kpc (blue) and central 0.4 kpc (in diameter, red). We mark averages (log10(mean(line/CO))) for the full FoV and central apertures (dashed gray, blue, and red line, respectively). The histograms are normalized to have an integrated area of unity.

In the text

Fig. 8. Comparison with HCN observations from Querejeta et al. (2019) (Q19). Left panel: SWAN HCN moment-0 map with contours (∼5, 50, 200σ with σ the average noise) of Q19 HCN maps. Both maps are at 3″ resolution, spatially and spectrally regridded to the same grid, and the maps were created via the same GILDAS island method (using just the HCN line). Circles depict radii of 35″ centered according to Q19. Right panels: Average spectra of SWAN 3″ data at 10 km s−1 resolution as well as the matched Q19 data. Spectra are the average flux inside of the circular area shown in the left panel.

In the text

	Fig. 9. Integrated emission maps (moment-0) from EMPIRE at 33″ for HCN, HCO+, HNC (1−0) (left columns) as well as from CLAWS at 15−30″ resolution for C18O, 13CO and N2H+ (1−0) (right columns) with SWAN 1, 5, 10 and 15 K km s−1 contours at native resolution (∼3″) on top. Beam sizes of EMPIRE/CLAWS and SWAN are shown in the bottom left corner of each panel (white and red circles, respectively).
In the text

Fig. 10. Literature comparison of integrated intensities for the J = 1 − 0 transition of (from top to bottom): HCN, HNC, HCO+, and N2H+ emission compared to 12CO. We show the SWAN integrated emission (light-shaded area) for the full FoV (squares), the central 1 kpc (triangle), and the remaining disk (circle). These values are obtained by integrating emission from pixels where both the line emission and 12CO emission is significantly detected (> 3σ). The distribution of pixels in the full FoV is added as violins. The error bars correspond to the difference between this calculation (both 12CO and the line are significantly detected) and when instead using pixels where 12CO is significantly detected. Average literature values (horizontal dashed lines, calculated on a linear scale) are based on values from Milky Way (Pety et al. 2017; Barnes et al. 2020; Jones et al. 2012) and extragalactic sources: M51 studies at lower resolution (Watanabe et al. 2014) and in the outer spiral arm (Chen et al. 2017); different galaxy averages from EMPIRE (Jiménez-Donaire et al. 2019); ∼100 pc studies in M33 (Buchbender et al. 2013), M31 (Brouillet et al. 2005) and NGC6946 (Eibensteiner et al. 2022); ∼10 pc observation in the LMC (Nishimura et al. 2016a) and 80 pc in the dwarf galaxy IC10 (Nishimura et al. 2016b); ∼15 − 19″ in three nearby galaxies (Takano et al. 2019) and upper limits from z ∼ 3 galaxies (Rybak et al. 2022). The values are basically sorted by their physical resolution from a few parsecs in the Milky Way (left) to several kiloparsecs in external galaxies (right).

In the text

Fig. 11. Same as Fig. 10, but for the J = 1 − 0 transition of the CO isotopologs C18O, and 13CO, as well as the HNCO(5−4) and HNCO(4−3) lines. The EMPIRE isotopologs are measured by Cormier et al. (2018) instead of the survey paper. Further, we add CLAWS measurements of M51 (den Brok et al. 2022). The measurements are sorted by physical resolution (increasing to ∼kiloparsec scales at the right side of the plot).

In the text

Fig. A.1. Quantifying errorbeam contributions to the observed 13CO(1–0) emission. (Left) The map presents the percentage errorbeam contribution to individual sightlines for the 30m only data. We only consider sightlines where the integrated 13CO(1–0) brightness temperature is detected at ≳3σ. Contours indicate the S/N at 5,10,20, and 50. The red contour illustrates the SWAN FoV. Overall, the error beam contribution is marginal, with elevated values up to 10% within the interarm region (and < 5% within the SWAN FoV). (Right) We extract 13CO(1–0) spectra in six individual apertures. Apertures 1–4 are in the interarm region and apertures 5–6 in the spiral arms of M51. Each panel shows the full spectrum which includes the contribution from the error beam in blue. The brown spectrum represents the error beam contribution and was calculated by subtracting the error beam free spectrum from the observed spectrum. The percentage contribution is computed over the spectral range where we detect emission (indicated by the blue-shaded region) and is listed in each panel.

In the text

	Fig. B.1. Pixel-by-pixel comparison of the HCN (1-0), HCO+ (1-0), and HNC (1-0) data cubes obtained with SWAN and EMPIRE IRAM-30m observations. The red line indicates a 1-1 correlation between the EMPIRE and the SWAN datasets. Both single-dish datasets are processed with the EMPIRE pipeline for comparison and overall agree with each other.
In the text

Fig. C.1. Integrated intensity maps (moment-0) of the SWAN dataset (combined NOEMA and IRAM 30m observations) for the J = 1–0 transitions of 13CO, C18O, N2H+, HCO+, HNC, HCN, C2H(1–0 and HNCO(J = 4–3) plus HNCO(J = 5–4) at their native angular resolution (∼2.3 − 3.1″). The maps were created with the GILDAS Island-method. We show the beam size in the bottom of all panels as well as a 1 kpc scale bar in the top left panel.

In the text

Fig. C.2. Integrated intensity maps (moment-0) of the SWAN dataset (combined NOEMA and IRAM 30m observations) for all detected lines at a common resolution of 3″ (125 pc). The data is binned with hexagonal spacing with the PyStructure code (den Brok et al. 2022; Neumann et al. 2023). Spectral windows for the creation of the moment maps are selected based on significant detections of 12CO emission from PAWS (Schinnerer et al. 2013) and HCN(1-0) emission. We show the beam size, a 1 kpc scale bar and mark the central 1 kpc circular area (green points) in the top left panel. The intensity scale is the same as in Fig. C.1.

In the text

Fig. C.3. Pixel-by-pixel comparison of the obtained integrated line emission using two different methods for the 10 km s−1 resolution 13CO data cube at native angular resolution. We show pixels located inside (black circles) and outside (gray circles) the hull of the mosaics (compare this with Fig. 1). We show the 1:1 relation (dashed orange line). We mark the average 5σ noise level for both lines (dashed gray line). We note that due to logarithmic spacing, data points containing noise with negative fluxes are not visible. Although this applies to most data points in the interarm region near the edges of our FoV, we emphasize that this comparison is intended to assess how both methods handle regions with significant detected emission, as these areas are typically the focus of scientific analysis. Regions with significantly detected emission is found mostly in the center, the molecular ring, and on the spiral arms.

In the text

	Fig. E.1. Same as Fig. 7 but for line ratios with 13CO(1-0).
In the text