ALMA CO(2-1) observations in the XUV disk of M83

The extended ultraviolet (XUV) disk galaxies are one of the most interesting objects studied in the last few years. The UV emission, revealed by GALEX, extends well beyond the optical disk, after the drop of H$\alpha$ emission, the usual tracer of star formation. This shows that sporadic star formation can occur in a large fraction of the HI disk, at radii up to 3 or 4 times the optical radius. In most galaxies, these regions are poor in stars and dominated by under-recycled gas, therefore bear some similarity to early stages of spiral galaxies and high-redshift galaxies. One remarkable example is M83, a nearby galaxy with an extended UV disk reaching 2 times the optical radius. It offers the opportunity to search for the molecular gas and characterise the star formation in outer disk regions, traced by the UV emission. We obtained CO(2-1) observations with ALMA of a small region in a 1.5'$\times$ 3' rectangle located at $r_{gal}=7.85'$ over a bright UV region of M83. There is no CO detection, in spite of the abundance of HI gas, and the presence of young stars traced by their HII regions. Our spatial resolution (17pc x 13pc) was perfectly fitted to detect Giant Molecular Clouds (GMC), but none were detected. The corresponding upper limits occur in an SFR region of the Kennicutt-Schmidt diagram where dense molecular clouds are expected. Stacking our data over HI-rich regions, using the observed HI velocity, we obtain a tentative detection, corresponding to an H$_2$-to-HI mass ratio of $<$ 3 $\times$ 10$^{-2}$. A possible explanation is that the expected molecular clouds are CO-dark, because of the strong UV radiation field. The latter preferentially dissociates CO with respect to H$_2$, due to the small size of the star forming clumps in the outer regions of galaxies.


Introduction
Over the last decades XUV disk galaxies have gained interest. Where Gil de Paz et al. (2005) shows the presence of UV-bright complexes in the outermost of some galaxy disks. These are called XUV disk galaxies, hosting UV emission well beyond their optical radii. UV-bright disks extending up to 3 to 4 times their optical radius (R 25 ) have been reported in about 30% of spiral galaxies (Thilker et al. 2005a;Gil de Paz et al. 2007b). Their extended UV emission covers a significant fraction of the area detected in HI at 21 cm wavelength (Bigiel et al. 2010;Boissier et al. 2003;Zaritsky & Christlein 2007). Generally, the UV starformation (SF) is related with this extending HI structure, e.g. shows evidence for metal enrichment (Gil de Paz et al. 2007a). Their far ultraviolet (FUV) and near ultraviolet (NUV) colours are generally consistent with young populations of O and B stars which probe a wider range of ages than H α and at low SF levels the number of ionizing stars may be very small (Dessauges-Zavadsky et al. 2014;Boissier et al. 2003).
Studying star formation beyond the optical radius allows us to address the condition of low-metallicity environments (Bigiel et al. 2010) . The study of SF in nearby galaxies involves the star formation rate (SFR), its surface density (Σ S FR ) and the gas surface density (Σ gas ) including both molecular and atomic gas. The relation between these quantities is known as the Kennicutt-Schmidt (KS) relation (Σ S FR ∝ Σ n gas , (Schmidt 1959;Kennicutt 1998;Kennicutt & Evans 2012)). This relation describes how efficiently galaxies turn their gas into stars, in quantifying the star formation efficiency (SFE). The KS relation is almost linear when most of the gas is molecular, providing a constant gas consumption time-scale of about 3Gyr (e.g Bigiel et al. (2011); Saintonge et al. (2011)). The SFE falls very quickly when the Σ gas < 10M /pc 2 when the gas is mainly atomic. However, recent surveys of molecular gas at high resolution have the sensitivity to probe this relation at Σ gas 3M /pc 2 . Studies of SF beyond the optical radius primarily concentrated on comparing different SF tracers. Until now studies of SFE in XUV disk was principally focused on atomic gas. These environments are not propitious to H 2 formation due to the low gas density and low metallicity conditions, which are akin to galaxies in the early universe.
The confirmed occurrence of star formation in the outer disk of normal spirals has several important implications: it indicates the presence of molecular gas in the outskirts of spirals, possibly an efficient phase transition from HI to H 2 . It is one of the regions to study the unresolved issue of the atomic hydrogen gas origin: either HI is the main phase, that can be transformed to H 2 to form stars, or reversely, it is the product of the star formation process, i.e. the result of the photodissociation of H 2 by the UV flux radiation emanated from newly formed stars (Allen et al. 2004;Smith et al. 2000). It also provides a simplified laboratory Article number, page 1 of 7 arXiv:1807.01373v1 [astro-ph.GA] 3 Jul 2018 for determining the star formation threshold. It is also the place to investigate the star formation in quiescent and low-metallicity environments that may affect the SFE and the initial mass function. Crosthwaite et al. (2002) studied the overall molecular gas morphology of M83 with CO(1-0) and CO(2-1) covering the 14'x14' optical disk.In this study they found that CO falls rapidly at radius of 5'=7kpc, and they proposed that this is also the decline in the total gas surface density, even if the HI emission continues further out, but with a lower surface density. Molecular gas dominates inside 7kpc radius (80% of the gas), while in total it is only 30% of the gas. The CO(2-1)/CO(1-0) intensity ratio is ∼1. At 7kpc, where the disk begins to warp, the ISM pressure might reach a threshold for the formation of molecular clouds. Thilker et al. (2005b) with GALEX observations have modified this view: diffuse UV emission is detected beyond the bright star-forming disk, when Hα and CO emission drop. This discovery made M83 the prototype of XUV disk galaxies. Koda et al. (2012) report deep Subaru Hα observations of the extended ultraviolet disk of M83, and found some weak emission, not seen by Thilker et al. (2005b). Dong et al. (2008) with S PIT ZER show that the SF has been an ongoing process in the extreme outer parts of M83 for at least 1Gyr. In a comparison between HI and FUV emission, Bigiel et al. (2010) show that the most extended atomic gas observed in M83 will not be consumed by insitu SF, and this might be due to the low efficiency of the HIto-H 2 phase transition there. However the present low SF might be sufficient for chemical enrichment. A flat oxygen abundance gradient was obtained beyond R 25 by Bresolin et al. (2009): they find only a slight decline in abundance beyond this galoctocentric distance with 12+log(O/H) between 8.2 and 8.6. Bresolin et al. (2016) present a chemical evolution model to reproduce the radial abundance gradient of M83 in R 25 , and their model is able to quantify the metallicity of the gas, which is very close to that of the stars.
Star formation in low gas surface density, less than 10 M pc −2 , is not very well known. The SFE in these regions is very low and seems to be uncorrelated with Σ gas : SFR has a much larger dynamical range than the local surface density (Boissier et al. 2007). CO observations in such environments are rare, due the weakness of the emission. A robust, quantitative picture of how the environment in the outer disks affects star formation is crucial to understand the origins of galaxy structure. Dessauges-Zavadsky et al. (2014) were the first to study molecular SFE in an XUV disk, M63, where they detected CO(1-0) in 2 of 12 pointings using the IRAM-30m. The authors concluded that the molecular gas in those regions has low SFE compared to regions in the inner disc. There are only four galaxies with molecular detections in the outermost disks, beyond R 25 : in addition to M63, NGC4414 (Braine & Herpin 2004), NGC 6946 ), M33 (Gratier et al. 2010). NGC4625 has been actively searched for CO emission but not detected (Watson et al. 2016). In these papers, the SFE is defined as Σ S FR /Σ H 2 , and we adopt this definition here (Dessauges-Zavadsky et al. 2014).
In this paper, we present ALMA CO(2-1) data covering one region outside the optical disk of M83. We describe our ALMA observations in more details in Section 2, and our results in section 3. Section 4 presents the discussion about the SFR in low environments and physical reasons to explain the dearth of CO emission in the outer M83 disk.

Observations
We observe CO(2-1) emission at 229.67 GHz (band 6) in M83 with ALMA during Cycle 2 (PI: Monica Rubio). The selected region is located at 13h37m03.6s − 29d59 47.6 enclosed in a 3 × 1.5 (4kpc × 2kpc) rectangle located at r gal = 7.85 = 11kpc from M83 center. We used the selection criteria corresponding to FUV/NUV GALEX images (Gil de Paz et al. 2007a), using the peaks of UV emission, as well as the correlation with the HI emission from the THINGS survey . By choosing these peaks of emission we focused on the outer parts of the UV disks, beyond the r 25 optical radius, where we are interested in detecting the molecular gas (see figure 1).
These selection criteria were used for the M63 XUV disk (NGC 5055), and led to a CO detection (Dessauges-Zavadsky et al. 2014) far outside the r 25 limit, while Schruba et al. (2011) only obtained an upper limit at 300" from the galactic center. The selected region of M83 was observed during 1hour in March 2014, in very good weather conditions (pwv 1.3mm). The 12m array was used with 34 antennas and a maximum baseline of 558.2 m. The map was done with a mosaic of 121 pointings separated by 12.9" (figure 2), and with an integration time of 10.8 sec per pointing.
The data were calibrated using the CASA reduction package. Approximately 36% of the data was flagged which was current at this epoch for 12m Array data. We produced a CO(2-1) data cube for each pointing with natural weighting, and for a velocity range from 15 to 1015 km/s, a channel spacing of 2.5 km/s and an rms of 10.3 mJy per channel. The calibrated 121 uv-tables were subsequently exported to GILDAS where the cleaning and the cube analysis were performed.
As for the CO line, We can estimate the upper limits found in each beam, assuming a profile width of 15 km/s FWHM. The rms noise level is 4mJy in 15km/s channels for each beam of 0.75"x0.56". The 3σ upper limit of the integrated emission is therefore 0.180 Jy km/s. This corresponds to L'CO(2-1) = 2.5 10 3 K km/s pc 2 . Assuming an intensity ratio of I 21 /I 10 0.7 (e.g. Braine & Combes (1992)), and a standard CO-to-H 2 conversion factor α = 4.36M /(Kkm/spc 2 ), this corresponds to M(H 2 ) = 1.510 4 M . This mass scale is much lower than a Giant Molecular Cloud (GMC) in the beam of 17 x 13pc.
Another way to see this limit is to consider that our 10.3mJy/beam noise level in 2.5km/s channels corresponds to a brightness temperature of 0.55K. Therefore clouds of 1.6K should have been detected at 3σ. In the outer parts of the Milky Way, as far as R=20kpc from the Galactic center, molecular clouds up to 10K (in CO(1-0)) and 5K (in CO(2-1)) have been observed, at similar widths (e.g. Digel et al. 1994;Sun et al. 2015Sun et al. , 2017. They would have been detected quite easily in our survey. . The HI emission is from the THINGS survey. The yellow ellipse shows the optical radius (r 25 = 6.09 = 8.64kpc) and the rectangle marks the region observed with ALMA and analyzed in this paper, located at r gal = 7.85 = 11kpc.

Results
We searched in an automatic way for CO (2-1) emission with an SNR of more than 5 σ in each of the 121 cubes, with the "detection assessment" procedure in GILDAS. We found 14 possibilities of emission, falling in the HI range of velocities (500-700km/s) in this region of the galaxy, out of the 121 cubes. However, there is no spatial coincidence between these hints of emission and any of the other tracers. Also the width of the profiles are in general too large. Therefore evidence for the CO emission from molecular clouds is very weak.
Reversely, we look for the most prominent clumps from other tracers, in particular Hα. Using the observations from the Subaru telescope (Koda et al. 2012) we found 13 regions of star formation, and tried to search for CO emission there. These regions are shown in the figure 3, where the circles scheme the spatial regions correspondant to our choice. After looking to the ALMA data, none of the hints found have more than 3σ signal.

Matched Filter Technique
The reliable detection of weak signals is an issue in astronomical data. The data analysis process can lead us to underestimate the probability of false detection, that is why it is necessary to estimate it. A technique that is expected to have among the best probability of true detection is the Matched Filter (MF) technique. The goal of the filter is to maximise the detectability of signal of known structure inside a random noise Gaussian (Vio & Andreani 2016). To better comprehend the quality of our observations (e.g. to find out whether our measurements are biased) we first apply a simple technique. Here we want to check the assumption that the probability density function (PDF) of the noise peaks is close to a Gaussian. In Vio & Andreani (2016), they show from a zero-mean map that when the positions and number of sources are unknown the matched filter could underestimate the probability of false detections. We use here the simplest technique, where we plotted the pixels values from our moment 0 data. The values are plotted in figure 4. The plot shows no obvious irregularity or departure from a Gaussian, revealing no problem with the cleaning or reduction.
Since we know that the signal should correlate to the other ISM tracer, the HI-21cm emission, which gives us the spectral region to find the line, we now apply this filter to the data,through a stacking technique in the next section.

Stacking of CO spectra, according to the HI velocity
The CO emission is normally associated with HI peaks, because the host molecular clouds probably form in regions of relatively    4. Histogram of the pixels values from the ALMA data cube (our observations). The blue histogram comes from the pixel values and the red curve is a Gaussian fit. We see that the noise is a perfect gaussian which does not show any indication of some bump that could be related to the reduction and deconvolution procedures.
high HI column density, in absence of spiral density waves Gardan et al. (2007) . Therefore, to exploit the full mosaic of data from ALMA, we must stack all individual spectra where emission is expected, at the HI expected velocity. Averaging all pixels of the map has not given any signal; this might be due to velocity dilution, since there is a significant velocity gradient, of ∼ 100km/s over the corresponding HI map (cf Figure 5 at right). For this CO averaged spectrum, we find an rms of 0.017 mJy in 14km/s channels. We therefore computed the first moment of the HI cube, to have an average velocity at each 1.5 x 1.5 pixel of the map. We smoothed the CO(2-1) map at the 1.5 resolution, and for each beam, we shifted the spectra of the right velocity amount to have all of them centered at the expected velocity, known from the HI spectrum. Then we averaged all spectra, where the HI integrated intensity was above a certain threshold, which was 50 mJy/beam × km/s. The same stacking was done on the HI cube, which gives the result plotted in Figure 5 at left. The resulting HI stacked spectrum has a FWHM of 33 km/s. No baseline was subtracted to the stacked spectra. The stacked CO(2-1) spectrum shows a hint of emission (at 3.5σ) with a FWHM of 14 km/s. When properly reduced, the average integrated flux is 0.85 mJy km/s. Adopting the conversion factor described in section 2, this low value corresponds to an average mass of 65 M of molecular gas per beam, but spread over 3 x10 4 beams, therefore would represent a total mass of 2 x 10 6 M over the whole mapped region, of 4 x 2 kpc. The same computation can be done on the averaged stacked HI spectrum. With a flux of 0.1 Jy/beam km/s, and a beam of 15.2 x 11.4 , this gives a total HI mass of 7.0 x 10 7 M over the same region. The mass ratio between the molecular and atomic gas is then M(H 2 ) /M(HI) = 3 × 10 −2 or below.
Let us emphasize that the stacking technique is not used here to smooth out possible extended signal on scales that, anyhow, would be filtered out by the interferometer. On the contrary, we here aim at conserving our spatial resolution of 17pc x 13pc fitted to GMC scale, in order to avoid dilution of the cloud emission. The stacking, as usual, is averaging out several realizations of possible cloud emission.

Discussion
The extended ultraviolet disks, present in 10% of nearby galaxies, offer the opportunity to study the interstellar medium and star formation in extreme conditions with low average gas density and surprisingly abundant star formation. M83 was one of the first XUV detection in its outskirt regions and it is the prototype for these type of galaxies. However, no highly significant CO emission was detected in Cycle 2 ALMA maps. To analyse the impact of our observations in the context of star formation in outer XUV disks, we investigate the behaviour in the Kennicutt-Schmidt diagram of the tentative CO detections, in plotting the equivalent molecular gas surface density, and star formation surface density.

Star formation diagram
The SFR surface density can be determined from a combination of FUV and 24µm fluxes, with time-scale 10-100 Myr, using the calibration from : The Hα luminosity was corrected for internal extinction, through L(Hα) corr = L(Hα) obs + 0.020L(24µm), according to Kennicutt & Evans (2012). Over an aperture of 200 pc size around the two strongest HII regions, we find an SFR of 3 × 10 −4 and 3.6 × 10 −4 M yr −1 . These values are comparable to what is obtained from equation 1. The upper limits of CO emission are also averaged over the same area (200 pc size), to compute the corresponding surface densities. Schruba et al. (2010) have studied the dependence of the molecular depletion time in M33 on the scale considered, and they conclude that ∼ 300pc is the limiting scale below which the SF law as a function of gas surface density is likely to break. We consider a very similar scale, where the SF law should be relevant. We used the following equation to calculate the molecular hydrogen surface density: where I 1−0 is the CO(1-0) line intensity in Kkms −1 , and we assume that the CO(2-1) to CO(1-0) intensity ratio is R 21 = I 21 /I 10 = 0.7. With the spatial resolution of our observations, the flux scale is 0.018 Jy per K. The standard CO-to-H 2 conversion ratio of X CO = 2 10 20 cm −2 /(K km s −1 ) is adopted, and the number in equation 3 includes helium correction. We plot the Kennicutt-Schmidt relation in figure 6, where we compare all molecular data obtained in outer disks up to now, with the large sample of nearby galaxies from Bigiel et al. (2008). The total gas surface density Σ H 2 + Σ HI is completely dominated by molecular gas above 9 M /pc 2 , and the SFR relation can then be considered as linear with the gas surface density.
For M83, the regions used for this plot are the most prominent star forming regions in the Hα map. We have smoothed the molecular gas surface density over the extent of the Hα region size, about 200 pc, which leads to lower upper limits. In those regions, the average SFR surface density is high enough that we could expect to find only molecular gas. We then plot only the upper limit on H 2 surface density obtained by averaging over Fig. 6. Kennicutt-Schmidt diagram relating the SFR surface density to the molecular gas surface density, adapted from Verdugo et al. (2015) and Bigiel et al. (2008). Dashed ovals represent the data from the outer parts of XUV disk galaxies: NGC4625 and NGC6946 from Watson et al. (2016), and M63 (NGC5055) from Dessauges-Zavadsky et al. (2014), taking only the molecular gas into account in all of them. The 3 green vertical bars are from NGC4414 (Braine & Herpin 2004), and the black circles for M33 (Gratier et al. 2010). The dashed vertical line corresponds to 3 M /pc 2 , the sensitivity limit of the CO data in Bigiel et al. (2008). Dashed inclined lines correspond to depletion times of 10 8 , 10 9 and 10 10 years to consume all the gas at the present SFR. The horizontal upper limits correspond to our CO(2-1) results on the M83 XUV disk of Σ H 2 , in two of the main SFR regions, of sizes 200 pc. the 200 pc-sized region. For all other compared galaxies, only the molecular component is taken into account. It can be seen that the time to consume the molecular gas in the outer regions of M83 is smaller (t dep < 3 10 8 yr) then the depletion time in nearby galaxies (t dep = 3 10 9 yr). It is not relevant to represent in this diagram the average value obtained through the stacking over the whole region of 4kpc x 2kpc, since the surface densities are then diluted to a region much larger than the usual star forming regions, used in this diagram (about 200-500 pc in size). It is indeed over such scales that a star forming region can be defined. At lower scales, the star formation tracers, such as Hα or FUV, are not expected to be correlated to the molecular clouds of their birth, since newly born stars progressively drift away at a different velocity than the dissipative gas. For the other galaxies, the gas and SFR surface densities concern also such scales, like M63 (Dessauges-Zavadsky et al. 2014) corresponding to the IRAM-30m telescope beam of 0.5 and 1kpc in CO(2-1) and CO(1-0) respectively.

The dearth of CO emission in the outer disc of M83
One main issue to explain the absence of CO emission in galaxies is the metallicity Z of the gas. It is now well established that CO emission is frequently very faint or nonexistent in lowmetallicity galaxies, for example in gas-rich irregular galaxies Elmegreen et al. (1980); Tacconi & Young (1987); Taylor et al. (1998). Both observations and theory indicate that the CO-to-H 2 conversion factor, X CO , increases at low metallicity below 12 + log(O/H) = 8.2 − 8.4 Bolatto et al. (2013). The dependence of this relation is non-linear with Z, and may be in Z −2 or steeper at low Z, because of UV photo-dissociation in the absence of dust, in addition to the under-abundance of CO.
However, in the M83 field observed with ALMA, there is only little abundance deficiency, the average metallicity of the H II regions is 12 + log(O/H) = 8.4 Bresolin et al. (2009), i.e. 1/2 solar. With a slight change of X CO for a higher value than standard, we should still be able to detect molecular clouds at more than 8σ. This means that the scatter in the X CO -metallicity relation could not explain the dearth of CO emission.
The formation of the CO molecule occurs from OH via ionneutral reactions that form HCO+, which after its dissociation forms CO. Thus, the CO formation rate depends on the abundance of OH, itself related to the rate of destruction by UV photons. The destruction of the CO molecule is mainly by photodissociation. In diffuse environments, the molecular clouds are more isolated, and less shielded. A schematic view of a molecular cloud includes a dense core, with CO emission, and a surface where CO is photodissociated, and emitting essentially in C+ and C lines (see figure 8 in Bolatto et al. (2013)). Because H 2 and CO have similar dissociation processes in their line-transitions, H 2 can partially shield the CO molecule, but with large CO column densities, the CO molecule is self-shielding. At low CO column densities, and in particular in the outer parts of molecular clouds, the CO molecule is absent, and the CO/H 2 abundance ratio drops. All the carbon is found in C and C+, and this gas is called CO-dark molecular gas.
Another possible scenario which could explain the lack of CO emission is that the photo-dissociated region is fragmented in smaller clouds. Each one has not enough depth to have CO emission, and is dominated by C+ emission. In outer disks, the small star forming clouds will emit much less together, for the same amount of gas, than the larger clouds present within the optical disk. The same scenario was invoked in Lamarche et al. (2017) where they failed to detect CO emission towards the starbursting radio source 3C368 at z=1.13 using ALMA data. They discuss the possibility that they observe far-infrared finestructure oxygen lines in star-forming gas clouds before they have had the chance to form an appreciable amount of CO. Watson et al. (2016) did not have a precise reason for not detecting CO in a young star-forming region of NGC4625, while Braine et al. (2007), Braine & Herpin (2004) and Dessauges-Zavadsky et al. (2014) could detect weak CO emission in the outer parts of NGC 6946, NGC 4414 and M63, respectively. However, in M63 CO is detected only in two out of 12 XUV-disc regions. Watson et al. (2016) conclude that, even if one explanation for the dearth of CO emission in lower-mass galaxies can still be the low metallicity, the issue remains for the outskirts of massive galaxies. Deeper observations are needed to disentangle the various proposed scenarios.
The gas metallicity is only half solar in our mapped ALMA field, which could already reduce the size of the CO clouds with respect to H 2 clouds. The dearth of CO emission could come also from the excessive local FUV radiation field, which dissociates CO preferentially, and from the small size of the clouds in the outer regions of some galaxies. M83 shows particularly strong FUV emission in its outer regions, with respect to others galaxies: the FUV flux ranges between 0.2 and 1.2 × 10 −2 MJy/sr over our region. The resolution of the FUV observations is 100 pc, at the distance of M83, so we cannot know exactly the radiation felt by each cloud. Although the FUV flux is much smaller than in the inner galaxy disk, clouds are smaller and less numerous in outer parts of galaxies (here at half metallicity), which explains the more extended CO dissociation. This could be explored further through follow-up observations of the dust continuum in the Rayleigh-Jeans domain in the same region, another independent tracer of the molecular gas.

Conclusions
We have reported about a mosaic of CO(2-1) observations obtained with ALMA in the M83 outer disk, rich in atomic gas and UV emission. M83 is the prototype of spiral galaxies with an extended XUV disk.
Our aims were to map CO emission in a small region r gal = 11kpc from the galaxy center. The result is a dearth of CO emission, leading to the following conclusions: 1. An automatic search of CO emission in the data cube provides tentative detections of 14 clouds, but the CO(2-1) signal at 4-5 σ does not correspond spatially and spectrally to any other tracer. They are therefore considered as false detections. 2. Reversely, we searched the CO map and extract the spectra at the peak of the star formation regions traced by Hα. This did not provide any CO detection higher than 3σ. 3. We have stacked all the CO pixels, at the places where significant HI signal is found, shifting their velocity scale to a common central velocity, expected from the HI signal. This gives a hint of detection with a profile width of ∆V = 14 km/s. The corresponding H 2 mass all over the 4 x 2 kpc area is only 2 x 10 6 M . The H 2 -to-HI mass ratio over this region is < 3 × 10 −2 . 4. We display the CO upper limits towards the star forming regions in the Kennicutt-Schmidt diagram, and the depletion time to consume the molecular gas is lower (< 3 x 10 8 yr) than in normal galaxy disks (3 x 10 9 yr). We could have expected to find in some pixels of 17 x 13pc GMC masses of 10 6 M , while the 3σ upper limits are ∼ 10 4 M . 5. The explanation for this lack of CO emission could be due partly to a low metallicity, since the gas abundance in this region of M83 is half solar. The average metallicity of the H II regions is 12 + log(O/H) = 8.4 Bresolin et al. (2009). Other causes of the dearth in CO may be the strong UV field, and low global density of gas and dust; the gas is predominantly H 2 but most carbon is not in CO molecules. In the photodissociation regions, the carbon is in C and C+. 6. In the outer parts of galaxies, at low gas surface density, the size of the clouds are likely to be smaller than in the disk, and less self-shielded. The CO column density in each cloud is then not sufficient to avoid dissociation, and the region is dominated by C+ emission.