Herschel -HOBYS study of the earliest phases of high-mass star formation in NGC 6357

Aims. To constrain models of high-mass star formation it is important to identify the massive dense cores (MDCs) that are able to form high-mass star(s). This is one of the purposes of the Herschel /HOBYS key programme. Here, we carry out the census and characterise of the properties of the MDCs population of the NGC 6357 H II region. Methods. Our study is based on the Herschel /PACS and SPIRE 70 − 500 µ m images of NGC 6357 complemented with (sub-)millimetre and mid-infrared data. We followed the procedure established by the Herschel /HOBYS consortium to extract ∼ 0.1pc massive dense cores using the getsources software. We estimated their physical parameters (temperatures, masses, luminosities) from spectral energy distribution (SED) ﬁtting. Results. We obtain a complete census of 23 massive dense cores, amongst which one is found to be IR-quiet and twelve are starless, representing very early stages of the star-formation process. Focussing on the starless MDCs, we have considered their evolutionary status, and suggest that only ﬁve of them are likely to form a high-mass star. Conclusions. We ﬁnd that, contrarily to the case in NGC 6334, the NGC 6357 region does not exhibit any ridge or hub features that are believed to be crucial to the massive star formation process. This study adds support for an empirical model in which massive dense cores and protostars simultaneously accrete mass from the surrounding ﬁlaments. In addition, the massive star formation in NGC 6357 seems to have stopped and the hottest stars in Pismis 24 have disrupted the ﬁlaments. are


Introduction
High-mass stars (O-B3 type, >8 M ) are the ionising sources of H II regions. They impact the interstellar medium mainly via their UV radiation, the dynamical expansion of their H II region Full Table C1, Tables C2-C5, the reduced Herschel FITS images and the column density FITS image are only available at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http:// cdsarc.u-strasbg.fr/viz-bin/qcat?J/A+A/625/A134 Herschel is an ESA space observatory with science instruments provided by European-led Principal Investigator consortia and with important participation from NASA. and supernova events. While their stellar and supernova phases are optically bright, their earliest phase of evolution occurs in cold massive dense cores (MDCs with sizes and volume densities of ∼0.1 pc and >10 5 cm −3 , respectively, see Motte et al. 2007), observable only in the far infra-red (FIR) and submillimetre (sub-mm) domains. Thanks to the development of FIR and sub-mm instrumentation, our understanding of highmass star formation processes become clearer, suggesting a much more dynamical process than the formation of low-mass stars such as accretion streams, associated with converging flows, and cloud hierarchical global collapse Csengeri et al. 2011;Peretto et al. 2014  Herschel-colour image of NGC 6357: 70 µm (red, resolution 5.9 ) and high-resolution column density (green, resolution 18.2 ). The coordinates are Galactic coordinates. The dashed yellow line outlines the Galactic latitude above which we assume the emission belongs to NGC 6357 and that we used to build the HOBYS catalogue. The cyan and magenta symbols indicate the position of the OB clusters Pismis 24 and AH03J1725-34.4, respectively. Tigé et al. (2017), studying massive dense cores (MDCs) in the star-forming region NGC 6334, favour a scenario wherein ridges/hubs, MDCs and high-mass protostellar embryos form and grow simultaneously (see the review of Motte et al. 2018a). A few high-resolution studies have already been performed with (sub-)millimetre interferometers revealing that starless highmass cores are very difficult to find (e.g. Duarte-Cabral et al. 2013;Tan et al. 2013;Nony et al. 2018). Even in NGC 6334 no high-mass pre-stellar cores (corresponding to the high-mass analogues of low-mass pre-stellar cores) were found (Louvet et al. 2019) supporting this.
NGC 6334 has similar velocity and distance to NGC 6357 (the adopted distance is 1.75 kpc, Russeil et al. 2012) and since the extinction and the 1.2 mm emission morphology tend to indicate that they are connected by a filamentary structure (Russeil et al. 2010) we usually consider them as a "twin massive starforming complex". However, contrary to NGC 6357, NGC 6334 exhibits a dense molecular ridge and two hubs (Matthews et al. 2008;Tigé et al. 2017), meaning that it could have a different star-formation history. The main characteristic of NGC 6357 (l = 353.4 • , b = +0.6 • ) is its shell like morphology identified in Hα, and by its surrounding photo-dissociation region (PDR, as seen by its PAH-8 µm emission). Filled by hot gas (Cappa et al. 2011) the cavity is shaped by the ionising open cluster Pismis 24. A shell of fragmented molecular gas was also identified in CO emission (e.g. Massi et al. 1997) and Giannetti et al. (2012) show the presence of a velocity gradient in the region suggesting that the expansion of the ionised gas is pushing the molecular gas. In addition, at least four other H II regions (see Fig. 1 in Russeil et al. 2016), adjacent to the cavity, belong to NGC 6357, one of them being powered by the cluster AH03J1725-34.4 (Dias et al. 2002).
In this paper, we have focussed our study on NGC 6357 based on data (Herschel-HOBYS 1 imaging survey) and an approach and method similar to that presented by Tigé et al. (2017) for the region NGC 6334. Our main goal is to identify and characterise MDCs in order to compare the massive star-formation processes in both regions. Clouds are hierarchical multi-scale structures, which are sub-divided into 1 pc clumps, ∼0.1 pc MDC, and ∼0.01 pc cores. Given the Herschel angular resolution (12 at 160 µm) and the homogenous 1-3 kpc distances of HOBYS clouds, the HOBYS key programme is dedicated to identify and characterise 0.1 pc MDCs. Since density is a better criterion than mass to evaluate whether a cloud structure has the ability to form high-mass stars we chose to focus on the densest cloud structures. In this paper, we will then define MDCs as massive ∼0.03 to ∼0.3 pc cloud structures whose mass is defined in Sect. 4.1.
Combining the Herschel-HOBYS data with complementary images described in Sect. 2, we extract the dense cores and characterise their properties (Sect. 3). Section 4 presents a complete sample of MDCs with robust mass estimates. Finally, in Sect. 5, we discuss the MDCs properties in comparison with the ones belonging to NGC 6334. Conclusions are given in Sect. 6.

Observations
2.1. Herschel observations, data reduction, and column density images NGC 6357 has been observed by the Herschel space observatory with the PACS (Poglitsch et al. 2010) and SPIRE (Griffin et al. 2010) instruments 2 as part of the HOBYS ) Key Programme (OBSIDs: 1342204847 and 1342204848). Data were taken in five bands: 70 and 160 µm for PACS and 250, 350, and 500 µm for SPIRE with FWHM resolutions of 5.9 , 11.7 , 18.2 , 24.9 , and 36.3 , respectively. Observations were performed in parallel mode, using both instruments simultaneously, with a scanning speed of 20 s −1 . The size of the observed field is 1.7 • × 1.1 • , which corresponds to 52 pc × 34 pc at a distance of 1.75 kpc. Data were reduced using the Herschel Interactive Processing Environment (HIPE, Ott 2010) 3 software, version 10.0.2751. Versions 7.0 onwards contain a module which significantly removes the stripping effects that have been observed in SPIRE maps produced with previous HIPE versions. SPIRE nominal and orthogonal maps were separately processed and subsequently combined and reduced for de-stripping, relative gains, and colour correction with HIPE. PACS maps were reduced with HIPE up to Level 1 and, from there up to their final version (Level 3) using Scanamorphos v21.0 (Roussel 2013).
Column density maps were built both at the 36.3 and 18.2 resolutions of SPIRE 500 and 250 µm data. The procedure used to construct the 36.3 resolution image uses the SED (Spectral Energy Distribution) fitting method fully described in Hill et al. (2011Hill et al. ( , 2012. To build the high-resolution column density map a multi-scale decomposition of the imaging data was performed and described in detail in Appendix A of Palmeirim et al. (2013).
The dust opacity law used (κ 0 = 0.1 × (ν/1000 GHz) 2 cm 2 g −1 ) is similar to that of Hildebrand (1983) with β = 2 and assuming a gas-to-dust ratio of 100 (this dust opacity law is commonly adopted in the other HOBYS papers). Figures 1 and 2 present Hα, Spitzer 8 µm, Herschel/PACS 70 µm and column density images of NGC 6357. While the 8 (PAHs) and 70 µm (emitted from warm dust) emissions show large and extended filaments and delineate cavities, the Hα (ionised gas) emission suggests that they are filled by ionised gas. The column density map appears to be more patchy here than in the case of NGC 6334.

Ancillary data
We complemented our Herschel-HOBYS observations with submillimetre and mid-infrared data (see Table 1). The ATLASGAL survey 4 (APEX Telescope Large Area Survey of the GALaxy, Schuller et al. 2009), using the LABOCA/APEX camera at 870 µm, covered the NGC 6357 molecular cloud with 19.2 resolution. NGC 6357 was also covered with 24 angular resolution, by dedicated SIMBA/SEST 5 1.2 mm observations presented by Muñoz et al. (2007) and Russeil et al. (2010). We also used mid-infrared wavelength images from Spitzer/IRAC and MIPS at 3.6−24 µm, as part of the GLIMPSE (Benjamin et al. 2003) and MIPSGAL (Carey et al. 2009) surveys 6 .
NGC 6357 was also covered with surveys made by the WISE 7 and MSX 8 space observatories (Wright et al. 2010;Egan et al. 2003). WISE provides full-sky images at four mid-infrared bands, notably at 22 µm with an angular resolution of 12 . The Midcourse Space Experiment (MSX) surveyed the Galactic plane in four mid-infrared broad bands, including 21.3 µm, with 18.3 angular resolution. One should note that, due to its strong 5 SIMBA/SEST was a former bolometer array of the SEST 15 m. 6  In addition, we retrieved public JCMT-HARP-ACSIS 9 reduced datacubes of NGC 6357 (proposal M14AU32, PI J. Wouterloot). During this run, the central part of NGC 6357 (31.8 × 28.1 ) was observed at 345.795 GHz ( 12 CO(3-2)) and 330.587 GHz ( 13 CO(3-2)) and automatically reduced using ORAC-DR (Jenness et al. 2015). The 12 CO(3-2) and 13 CO(3-2) lines datacubes have respectively 1884 and 1852 channels with a velocity resolution of 0.42 and 0.44 km s −1 . The beam FWHM is 14 and the pixel scale is 7.27 pixel −1 . In parallel, we collected the 93 MALT90 10 datacubes pointing in the direction of NGC 6357. MALT90 (Jackson et al. 2013;Foster et al. 2013) is a survey of 2000 dense cores located in the Galactic plane led with the ATNF Mopra 22-m telescope. The database provides 3 ×3 datacubes for 16 lines simultaneously observed around 90 GHz with an angular and spectral resolution of 38 and 0.11 km s −1 , respectively. We complemented the above data with ATNF Mopra 22-m telescope observations (on the fly mapping) of the lines HCO + (89.188 GHz) and N 2 H + (93.174 GHz) observed in 2008 (reduced with Livedata and Gridzilla codes). The angular and spectral resolution are 30 and 0.11 km s −1 , respectively.
The detection module of getsources was run on the Herschel flux maps as well as on the high-resolution column density map. At this step the used PACS-160 µm and SPIRE-250 µm images are the temperature-corrected ones (see Tigé et al. 2017) helping getsources to identify the compact sources against the strong background emission. During this detection step getsources defines a catalogue of sources with a unique position.
For the measurement step, we used the original (not temperature-corrected) Herschel maps from 70 to 500 µm plus the available sub-millimetre maps listed in Sect. 2.2, that is, the 870 µm LABOCA and 1.2 mm SIMBA images. At this step, getsources derives flux measurements that are background subtracted and deblended from overlapping sources. The output source catalogue lists, among others, the monochromatic detection significance index 11 (Sig mono ), peak and integrated fluxes (with errors), FWHM major and minor sizes, and the position angle of the elliptical footprint for each extracted source in each far-infrared to sub-millimetre band.

Compact source selection
The Herschel-HOBYS imaging of NGC 6357 extends in Galactic latitude from approximately −0.4 • to +1.4 • (see Fig. 1). Russeil et al. (2016) note that sources below b ∼ +0.1 • (the yellow dashed line in Fig. 1) are Galactic plane sources not related to NGC 6357. By masking this area, we collected 922 (out of the 1391 sources located on the whole field) sources in the direction of NGC 6357 from the getsources catalogue.
In order to keep only reliable flux measurements and compact structures, and to be able to subsequently perform the SED fitting, we applied additional source selection criteria as described and discussed in Tigé et al. (2017). We summarise these selection criteria below.
-For each source and each band: -The signal-to-noise ratio (S/N) must be greater than two (limiting false detections while allowing the SEDs to be well constrained by many flux measurements) for both the peak and integrated fluxes 12 ; -The deconvolved size must be less than 0.3 pc (to discard clumps); -The aspect ratio must be smaller than two (to discard elongated, filament-like features). This first step selection process discards 41% of the sources. They are mainly sources with low signal-to-noise fluxes hence sources at the limit of the detection level in Herschel wavelengths and corresponding to dense cores with very low mass. We can note that only 12 (1.3%) sources have been discarded because of the size criterion.
-For the SED fitting we require a minimum of three reliable flux measurements: -One at either Herschel-160 µm or Herschel-250 µm, which we call the reference wavelength; -A second Herschel flux measurement at λ ≥ 250 µm; -A third flux measurement taken at λ > 250 µm with either Herschel-SPIRE, APEX/LABOCA, or SEST/SIMBA. From this second step selection process, 37% of the sources are discarded because they have no reliable fluxes at Herschel-160 µm and/or Herschel-250 µm. These sources are either PDR or elongated features. Indeed, dense cores should have SED peaking between 100 and 300 µm and should therefore have reliable flux measurements at Herschel-160 µm and/or Herschel-250 µm. Finally, 5% are excluded because they are 70 µm-only sources. This gives us a sub-sample of 155 robust sources which fullfill these selection criteria.

Compact sources physical characterisation
The SED fitting was made for the 155 selected sources to determine their mass and temperature (see Table C.1). To construct SEDs with fluxes measured within a similar aperture, we applied the flux scaling procedure and colour correction as described in Tigé et al. (2017). The study of the profile of each core is beyond the scope of this paper. As a result, a flux scaling was performed assuming that sources have a quasi-spherical radial density distribution following a ρ(r) ∼ r −2 law as it is observed for protostellar envelopes (e.g. Beuther et al. 2002;Nguyen Luong et al. 2011a). This flux scaling allows to reconcile Herschel fluxes with higher resolution observations and has been explained at length in for example Nguyen Luong et al. (2011a) and Tigé et al. (2017). We stress that it has a relatively weak impact (at most 50% decrease in mass) on the results.
The fitted model is a modified black-body model with a dust emissivity spectral index, β, set to 2 and the 70 µm flux is not used for the SED fitting except when an extreme temperature (larger than 32 K) is found (Tigé et al. 2017). Indeed, Men'shchikov (2016) demonstrated that mass derivation with a free variable spectral index leads to very strong biases and erroneous masses.
The flux uncertainties were evaluated as in Tigé et al. (2017) and the uncertainties produced by the fitting routine (similar to Tigé et al. 2017) allow us to obtain the mass and temperature uncertainties (their mean value being 35 and 11%, respectively). The core's mass and temperature are obtained by fitting the equation: where S ν is the dust continuum emission, d is the distance to the Sun, and κ ν the dust mass opacity (the chosen value of κ ν is discussed in Tigé et al. 2017). The volume-averaged density is then calculated (as in Tigé et al. 2017) by where µ = 2.8 and the size is either the deconvolved equivalent FWHM measured at the reference wavelength or, for sources unresolved at 160 or 250 µm, it is defined 13 as 0.5 × HPBW Refλ . Cross-matching (cone search 7 ) with GLIMPSE 3.6 − 8 µm, MIPSGAL 24 µm, WISE 22 µm, and MSX 21 µm catalogues 14 was performed to complement the SED in the mid-IR and near-IR ranges. In addition, counterpart with GLIMPSE sources with rising fluxes between 3.6 and 8 µm are favoured, since the 1 − 10 µm SED portion is expected to be rising for both Class I and II young stellar objects (YSOs) (Whitney et al. 2004;Molinari et al. 2008).
This allowed us to compute the bolometric luminosity (L bol ). We first computed the flux density integration over the finite number of reliable data-points (from the first available data-point  measurement to the last one in the 3.6 µm to 1.2 mm range) sampling the SED (L Data ) and using the trapezoid rule (requiring a linear interpolation in the flux density versus frequency space). The data-points which are upper limits are not used in this fitting process. In parallel, we calculated the flux integration in the same way but under the full fitted curve (L Fit ). Similarly, the submillimetre luminosity (L sub−mm ) is the flux integration under the fitted curve but only from 350 µm to 1.2 mm.
When no mid-IR and near-IR counterparts were found L Fit is a better evaluation of the true bolometric luminosity, then we adopted L bol = max (L Fit , L Data ) (in general L Fit ≥ L Data ). When mid-IR and/or near-IR counterparts are found the trapezoid integration method is favoured and then L bol = L Data . However, we must keep in mind that ∼19% of the sources fall in saturated areas of the 22 and 24 µm images and only five sources have a compact MSX 21 µm counterpart. When the 21/22/24 µm flux is missing this leads to an over-estimate of L Data , relative to the true bolometric luminosity, by a factor of between two and four (Tigé 2014). In addition, L Data must be considered with great caution when the source has near-IR counterparts but no 21/22/24 µm flux and no 70 µm flux. In this case the L Data is unrealistically large because the linear interpolation in the trapezoid integration method implies its strong over-estimation (by a factor of up to ten).
Due to its large uncertainty, L bol is used no further in this paper and does not affect its conclusions. Table 2 summarises the main properties of the sample of 155 cores and the mass distribution is shown in Fig. 3. The mass distribution (Fig. 3) peaking at ∼30 M , we adopted this value as completeness level.
allowed us to make an homogeneous comparison between the two regions. by looking for signposts of high-mass star formation. Nevertheless, to validate this choice we look for signs of high-mass star formation by cross-correlating (cone search 10 ) the 155 robust sources with Class II CH 3 OH masers (Caswell et al. 2010;Urquhart et al. 2013) and radio centimeter UCH II regions Condon et al. 1998;Giveon et al. 2005). We find maser or H II region association for only seven cores. However four cores have a methanol maser with velocity peaking at −51, −41 or −16 km s −1 respectively suggesting they do not belong to NGC 6357 (the systemic LSR velocity of the region is −4 km s −1 , Caswell & Haynes 1987) and two cores have compact radio counterpart, but their masses are low (M ≤ 10 M ) suggesting either that their association is erroneous or they are H II regions at different distances along the NGC 6357 line of sight. Only one core exhibits a CH 3 OH maser and a strong and compact MSX 21 µm emission (it is in a saturated area on MIPSGAL 24 µm and WISE 22 µm images). In addition this core is classified as an "extended green object" (an object within an enhanced Spitzer 4.5 µm emission usualy attributed to shock exited H 2 features tracing outflow, Cyganowski et al. 2008) by Chambers et al. (2014). Assuming the evolutionnary tracks from Molinari et al. (2008), the evolved nature of this core, with a present mass of 48 M , could suggest an earlier mass around 140 M consistent with the adopted mass limit. For NGC 6357, we finally obtain a selection of 23 MDCs. Table 3 gives their main physical properties while their fluxes, multi-wavelength images and SEDs are displayed in Appendixes C and D. Because the mass limit (75 M ) is well above the completeness level, we assume that we do not miss any MDC in NGC 6357. At a first look, we note that MDCs in NGC 6357 are slightly larger and less dense and massive than those in NGC 6334.

Nature and evolutionary status of MDCs
To estimate the evolutionary status of MDCs we followed a similar approach to that in Motte et al. (2007), Csengeri et al. (2014), König et al. (2017), Giannetti et al. (2017), and Tigé et al. (2017). We classified the MDCs into IR-bright protostellar MDC, IR-quiet protostellar MDC, starless MDC or undefined cloud. An MDC is qualified as an "IR-bright protostellar" if it has a 21, 22, or 24 µm flux larger than 10, 12, 15 Jy, respectively (for d = 1.75 kpc, Motte et al. 2010;Russeil et al. 2010), a maser and/or H II region (radio continuum) association and a clear centrally-located Spitzer-8 µm point source. But, because of the large area saturared at 22 and 24 µm and due to the poor coverage and quality of the radio and maser surveys it is difficult to find sources fullfilling all these criteria and we can miss IR-bright MDCs identification. Despite relaxing the IR-bright protostellar status to sources fullfilling at least one of these criteria we found no IR-bright protostellars MDC. As in Tigé et al. (2017), an MDC is qualified as an IR-quiet protostellar if it is associated with a 70 µm compact emission even if it is detected or not at 21/22/24 µm (but lower than 10/12/15 Jy).
An MDC is qualified as "starless" MDC candidate if no compact 70 µm and no 21/22/24 µm emission are detected, in addition to be centrally concentrated 15 If it is not centrally concentrated the MDC is then qualified as an "undefined cloud structure" (corresponding to unbound cloud structures). The qualification of undefined cloud structures was introduced by Tigé et al. (2017) and Rayner et al. (2017). These are cloud structures that are not centrally concentrated. To estimate the MDCs detection reliability we check whether they are also in the Hi-GAL (Molinari et al. 2016) source catalogue obtained with the CuTEx 16 algorithm. The two MDCs (see Table 4) straddling two Hi-GAL sources are undefined cloud structures.
The final sample of MDCs is plotted on Fig. 4. We note that they are located on high column density regions and they are mainly distributed along a strip close to l ∼ 353.1 • .

Velocity structure and massive dense cores properties
From the 12 CO(3-2) integrated profile a systemic velocity for the region of −3.5 km s −1 is measured, but several velocity components can be seen (Fig. 5). The −4 km s −1 component is well correlated with the column density, delineating the cavity that is clearly filled by ionised gas. In the direction of the cavity seen by Russeil et al. (2016), the ionised gas (Hα emission) shows a velocity of −2 km s −1 in the central part while the surroundings consist of an Hα semi-ring-like feature at −8 km s −1 , to which the region G353.  Notes. (a) The MDC's numbering is done by decreasing mass. (b) All MDCs are associated with a Hi-GAL source (Molinari et al. 2016) except when it is indicated as "No Hi-GAL cov". (not covered by the Hi-GAL survey) or "Over two Hi-GAL" (meaning that the MDC is straddling two Hi-GAL sources). (c) For MDC #23, the lack of mid-IR and 70 µm fluxes make a large and unreliable over-estimation of L bol . Then the quoted L bol is assigned to be L Data but calculated using the expected 70 µm flux evaluated from the fitted curve.
semi-continuous elliptical-like feature of 6.7 × 8.3 size, centred at l,b = 353.08 • ,+0.83 • (α, δ = 17h 24m 45.7s, −34 • 18 21.9 ), while it is found towards Hα extinction areas and superimposed on the −4 km s −1 emission around l,b = 353.15 • ,+0.67 • . Therefore, this position appears to be special due to the different velocity components mixing and extended profiles. This can be due to a combination of component superposition and self-absorption effects. However, this place corresponds to the contact zone between the cavity and the regions H II 353.09+0.63 and H II 353.24+0.60 (Russeil et al. 2016) where dynamical interaction of the ionised gas with the molecular cloud can occur. There are also young stars and OB stars Russeil et al. 2017) which can participate to the local turbulence by their feedback. Assuming that the −9 and the +2 km s −1 are the extreme velocities and −4 km s −1 the systemic velocity, we can estimate an expansion velocity of ∼5 km s −1 for the NGC 6357 central cavity. Finally, the patchy +6 km s −1 emission is usually attributed to a foreground layer .
We have extracted the spectra 17 at the position of 20 among 23 MDCs in order to measure the core's local environment velocity and to identify particular profile. The N 2 H + spectra 17 To increase the signal-to-noise ratio the spectra were extracted from 1.26 to 1 area for MALT90 and Mopra data, respectively. are fitted using the hyperfine fitting routine of the spectroscopic analysis toolkit "pyspeckit v0.1.20" (Ginsburg & Mirocha 2011) assuming a single temperature. However, for lot of N 2 H + spectra the modelled profile over-or under-estimate some of the hyperfine components suggesting that it is not consistent with a single temperature assumption. The HCO + spectra are fitted by simple Gaussian(s).
The optically thick HCO + line is usualy used to probe collapse or outflows (e.g. Fuller et al. 2005) as a double peaked profile with an excess of emission on the blue (red) side is generally interpreted as an indication of collapse or infall (outflow). In parallel, combining HCO + and N 2 H + , considered as optically thick and thin lines respectively, we can calculate the asymmetry parameter δv, as defined by Mardones et al. (1997), which indicates a clear asymmetry if | δv | > 0.25.
The results are listed in Table 5. The MDCs show a mean velocity of −3.74 km s −1 , while the MDC #23 has a velocity approximately −10 km s −1 (in agreement with its location in the pillar like feature previously noted at l,b = 353.16 • , +0.82 • ). For the IR-quiet protostellar MDC the spectral analysis is in agreement with its proto-stellar classification as it shows a double-peaked asymetric HCO + profile. However its HCO + blue and red peaks relative intensity suggest an outflow while the asymmetry factor suggests an infall motion. For starless MDCs,  four have | δv | > 0.25 while for undefined clouds none of them have a significant asymmetry. Interestingly, the starless MDC #23 has the most negative δv (suggesting collapse), and probably is in a slightly more evolved stage than the other starless MDCs.
To estimate the stability of the MDCs, virial parameter (defined as α vir = M vir /M MDC by for example, Bertoldi & McKee 1992) is calculated. For this, from the velocity width of the N 2 H + we derive the turbulent component as σ 2 turb = σ 2 N 2 H + − σ 2 th where σ th is calculated from the dust temperature. The virial mass and the virial parameter are calculated using σ turb . These quantities are reported in Table 5, for MDCs with no N 2 H + , σ turb and α vir are calculated from the HCO + line. The virial parameter can be used for estimating whether a structure is in a state of being gravitationally bound. In this way, we are able to probe the virial equilibrium for most of the MDCs finding that 14 of them have α vir ≤ 1, suggesting they are gravity dominated (and may be undergoing collapse), while four have 1 < α vir < 2 suggesting they are at least gravitationally bound and 1 MDCs has α vir > 2 suggesting it is unbound and may expand (in agreement with the fact that it is also classified as undefined cloud). On average undefined clouds structures have larger α vir but most of them have α vir < 2, suggesting they are not transient features.

Comparison with previous massive dense core studies in NGC 6357
We have compared our Herschel cores with the 74 clumps (∼0.4 pc) extracted at 1.2 mm by Russeil et al. (2010). We find that 36 clumps have single associated Herschel sources, 11 have two or three associated sources, and 27 have none. Of the 36 clumps associated with a single Herschel core, ten are associated with a M > 75 M cores 18 . Among the 11 clumps with two Herschel sources, one encompasses two massive cores, four encompass one massive core and six encompass low mass cores only. The 1.2 mm clumps with no associated Herschel source always have M < 75 M or a size larger than 0.3 pc. Inversely, 16 of the 23 MDCs are found in the direction of a 1.2 mm clump, 13 being located within a M ≥ 100 M clump. Giannetti et al. (2012) studied the molecular cloud associated with the region G353.2+0.9 facing Pismis 24. They distinguished, from molecular lines, 14 clumps (with size between 0.27 and 0.56 pc and a total mass of ∼2000 M ). From our sample, ten reliable cores are detected in the same area among which four are MDCs and appear located in the Giannetti et al. (2012) clumps of mass between 180 and 360 M . We also note that 18 of our 23 MDCs fall in an ATLASGAL compact source footprint Urquhart et al. 2018) and one (MDC #18) partly overlaps an ATLASGAL source.
All this highlights how this study allows us to see the fragmentation process within clumps. This is also in agreement with the results of Csengeri et al. (2017a) who find a limited fragmentation (at the typical scale of ∼0.06 pc) from a sample of 35 massive infrared quiet ATLASGAL clumps. This underlines the importance of the Herschel resolution and multi-wavelength information for the MDCs census and characterisation in comparison to previous studies.

Quest for the best birthplaces for high-mass stars
In NGC 6334, Tigé et al. (2017) identified 46 MDCs among which 16 are starless candidates and thus possible sites for the pre-stellar stage of high-mass star formation. However among these 16 starless candidates only one was held as the best candidate, while nine were excluded due to poor SED fitting, and the others due to their low density (1 − 5 × 10 5 cm −3 ) or because they were not centrally concentrated. Similarly in NGC 6357 we can evaluate the reliability of detected starless MDCs to be truly precursors of high-mass stars. In NGC 6357, except MDC #1, the SED fitting of the starless MDCs is good (χ 2 < 5). In addition, their mass is above the lower limit for high-mass star formation (given e.g. by the mass-radius relation M(r) ≥1282 × (r/pc) 1.42 M from Baldeschi et al. 2017) and their surface density, Σ, is larger than 0.05 g cm −2 (He et al. 2015) suggesting that all the starless MDCs are on first approximation possible sites of high-mass star formation. However, from mass-radius plots the massive-star formation limit is empirical, and based on fits to different clump or core samples. For example, Kauffmann & Pillai (2010) and Urquhart et al. (2014) find different slope and intercept values for their fitted power-laws than Baldeschi et al. (2017). In addition, as noted by Baldeschi et al. (2017) the massive star formation (MSF) thresholds certainly breaks at mass lower than 20 M . This is because the adopted values of the core-to-star conversion factor (between 0.5 and 0.33 from Alves et al. 2007) suggest it is not reasonable for a core (if there is no ambient accretion) that will form a high-mass star to have a mass lower than 20 M . Besides, some theoretical MSF thresholds suggest for radii smaller than 0.05 pc a minimum core mass between 50 and 90 M (e.g. see Fig. 1 in Kauffmann & Pillai 2010). In this context we need to precise the mass, density and size criteria to refine the number of starless MDCs able to form a high-mass star. Indeed, densities as low as n H 2 ∼ 10 4 -10 6 cm −3 is characteristic of low-mass pre-stellar cores (Könyves et al. 2015;Motte et al. 2007; Ward- Thompson et al. 1999) while small (<0.1 pc), high mass (M > 100 M ) and high density (n H 2 > 5 × 10 6 cm −3 ) are good criteria to select starless MDCs which could produce high-mass stars (e.g. Tigé et al. 2017;Urquhart et al. 2018). More generally, high resolution observations show that the dense (n H 2 > 10 5 cm −3 , compact ( 0.1 pc) and massive clumps tend to A134, page 10 of 42 be those that are fragmented and that contain the most massive cores Csengeri et al. 2017b;Figueira et al. 2018;Lu et al. 2018). It is the case for example of CygX-N63 MDC (which has n H 2 = 1.64 10 6 cm −3 ) and shows, at higher resolution, one massive fragment  as well the few massive protostellar cores of Lu et al. (2018) which have size ≤ 0.15 pc and n H 2 > 1.1 10 6 cm −3 . These characteristics are in agreement with the IR-quiet protostellar MDCs identified both in NGC 6357 and NGC 6334. For this reason, in this study, we have assumed that the compact (size ≤ 0.15 pc) and densest (n H 2 > 1 10 6 cm −3 ) starless MDCs are the best candidates to form a high-mass star. At this step, the starless MDCs #10, #11, #12, #17, #18, and #22 are the most favourable MDCs to form high-mass stars. To explore further the probability of MDCs containing high-mass starless cores or high-mass protostars we look at their mass concentration by plotting starless MDCs in the mass versus size diagram (Fig. 6). Motte et al. (2018a) and Tigé et al. (2017) recall that a mass concentration as Mass(<r) ∝ r is expected for ρ ∝ r −2 radial density structure (like it is observed for protostellar envelopes and the outskirt of pre-stellar cores) while Mass(<r) ∝ r 2 characterise much less concentrated clouds. Tigé et al. (2017) led such analysis for NGC 6334 starless MDCs (for which 40 and 10 resolution data exist) allowing them to finally select only one best starless MDC candidate. Following the same approach, in NGC 6357 (from Fig. 6), because they stray the most from the Mass(<r) ∝ r 2 law, MDCs #10, #11, #12, #17 and #22 appear more concentrated and then more favourable to form a high-mass star. We can note that these MDCs are mainly located in features facing the clusters Pismis 24 (MDCs #11 and #17) and AH03J1726-34.4 (MDCs #12 and #22).

Relation between MDCs and filaments
Recently Louvet et al. (2019), targeting with ALMA 13 of the 16 starless MDCs of NGC 6334, observed that four of them contain low-mass protostars while amid the others only one appears sub-structured into two low-mass pre-stellar cores. Even the best starless candidate in NGC 6334 (MDC-5) shows only low mass pre-stellar cores and protostars (Louvet et al., priv. comm.). This appears as a strong observationnal argument in agreement with the empirical model by Motte et al. (2018a)  This package allows us to identify spatially coherent structures and determine their morphological (like lengths and geometrical shape) and physical (e.g. column density, mass) parameters. Filament spine and branches are displayed on Fig. 7. Selecting structures with length-to-width ratio larger than two (see Appendix B) we plot their properties in Fig. 8.
The histograms (Fig. 8) show that the filament properties in NGC 6334 and NGC 6357 are statistically similar, but in NGC 6334 they have parameters reaching higher values. In Fig. 8 the histogram of the deconvolved widths is presented. However, due to the distance of the regions, the filament width is not resolved at Herschel/PACS 250 µm (used to produce our highresolution column density map) and are thus very uncertain. We do not use the filament width further in our analysis, their analysis is the purpose of the paper of Könyves et al. (in prep.) based on the 350 µm and 8 resolution map 20 .
The filamentary difference between the two regions becomes obvious on Fig. 7. NGC 6334 is dominated by a ridge corresponding to three aligned filaments with a total mass ∼31807 M filament with a mass per unit length (M line ) between 380 and 1893 M pc −1 , with a total length of 21.5 pc and a mean width of 0.08 pc. In NGC 6357 no ridge or hub is noted, the only distinct feature is the filament (composed of three segments) around l,b = 353.5 • , +0.66 • (hosting the IR-quiet MDC # 6) with a mass of 585.6 M (M line between 15 and 107 M pc −1 ), a total length of 8 pc and a mean width of 0.16 pc. Such difference in filament and MDCs concentration is already observed at different locations of other regions as, for example, Cygnus (Motte et al. 2007), in Lupus (Rygl et al. 2013) and Perseus (Sadavoy et al. 2014).
There is increasing observational evidence that filaments with M line larger (thermally supercritical) than the critical value M line,crit 21 show evidence of pre-stellar cores and YSOs whereas thermally sub-critical filaments (M line < M line,crit ) appear 19 The Vialactea Filamentary Structures Extraction Package is available at http://vialactea.iaps.inaf.it/vialactea/eng/ tools.php 20 APEX-ArtMiS, http://www.apex-telescope.org/ instruments/pi/artemis/, proposals E-094.C-0743 PI. P. André and O-094.F-9320 PI. T. Hill. 21 M line,crit is defined as M line,crit = 2 c 2 s G −1 ∼ 23 to 42 M pc −1 for a dust temperature of T = 14 and 25 K respectively (e.g. André et al. 2010  generally devoid of Herschel pre-stellar cores and protostars (André et al. 2010). This led to a proposed paradigm for solartype star formation in which low-mass stars form primarily by gravitational fragmentation of supercritical filaments . Hill et al. (2011) and Schisano et al. (2014) also suggest that the filamentary regions are more favourable to form massive stars and that MDCs are either in ridges or in hub within an A V > 100 mag environment (e.g. Nguyen Luong et al. 2011bLuong et al. , 2013Hill et al. 2011;Motte et al. 2018a). As highlighted by Li et al. (2016), the morphology of filaments varies from marginally resolved elongated structures to very complex networks of filaments and that they can be isolated or at the periphery of H II regions (as they are found by Inutsuka et al. 2015 to be preferential sites of filament formation). In addition, filaments that have high linear mass density (M line 100 M pc −1 ) contain large enough mass reservoirs to give birth to high-mass stars and star A134, page 12 of 42 D. Russeil et al.: Herschel-HOBYS study of the earliest phases of high-mass star formation in NGC 6357 5e19 1e20 2e20 5e20 1e21 2e21 5e21 1e22 2e22 5e22 1e23 2e23 Mean column density [cm ¡2 ]  to ∼2000 M pc −1 over nearly 10 pc and a width of 0.15 pc. A gap is noted around the middle of the NGC 6334 filament probably created by the H II region seen in Hα at this position and illustrating how recent massive star formation disrupts filaments. In addition, Zernickel et al. (2013) observe gas flow along the ridge which can be interpreted as another hint for the model by Motte et al. (2018a), model in which MDCs grow in mass from the surrounding gas. We clearly note that filaments more massive than 1000 M belong to NGC 6334 while the most massive filament in NGC 6357 reaches only 766 M . In both regions, MDCs are located on the filament, spine, or branch and they belong to the most massive and highest M line filaments (Fig. 8) indicating that filaments are important structures for the formation of massive stars. Among the five best candidates the MDCs #11 and #17 are located on a spine; the others are either on a branch or at the end of a small spine. Therefore, if we assumed that mass feeding along filaments is an effective process, MDC #17 could be the best to grow in mass as it is located at a branch to spine junction.
We can use the mass and the M line to estimate which of the starless MDCs in NGC 6357 has enough mass reservoir to form a massive star. The IR-quiet MDC (MDC # 6) belongs to a 5 pc long, 537 M , M line = 107 M pc −1 segment (while in NGC 6334 IR-quiet MDCs belong to filaments with M line between 224 M pc −1 and 1152 M pc −1 ). In this way, the starless MDCs #10, #11, #12, #17 and #22 are all in filaments that have M line > 107 M pc −1 (see Table 6). MDC #12 is quite peculiar because it is located at the end of a short (0.61 pc) filament (clump-like filament). Its velocity analysis (Sect. 4.3) indicates infall but the filament mass is similar to its mass suggesting that either the mass reservoir is not enough to form a massive star or that it is in a slightly more evolved stage than the other starless MDCs.
We also use class 0/I YSOs (from Povich et al. 2017) to probe the star formation activity of filaments where the starless MDCs are. Indeed, Rivilla et al. (2013Rivilla et al. ( , 2014 show that YSOs and low mass pre-main sequence stars tend to be clustered at the massive star-forming sites. They interpret that as evidence of the "competitive theory" for massive star-formation (e.g. Bonnell & Bate 2006), where a low-mass stellar cluster creates a potential well which funnels gas and dust towards its centre where the most massive stars will form. In this frame, we can speculate that MDCs not surrounded by YSOs could be less evolved or have less probability to form a massive star.
Several YSOs are identified around the MDCs #10, #12, and #22. Especially, around MDC#22 their spatial distribution seems to delineate an elongated structure (from l,b = 353.04 • , +0.57 • to l,b = 353.07 • , +0.40 • ) which encompasses three filaments (Fig. A.1  Notes. (a) Free-fall time (t ff ) measured from the median values of the density averaged over the full MDC volume, which is approximately a sphere with a FWHM radius: n H 2 full = n H 2 /8 and Due to the large uncertainty on L bol and any possible NIR source fortuitous association this value can reach a lower limit of 319 L (considering only that L bol = L Fit ). (Anderson et al. 2014) suggesting that massive star-formation is already processing at this place, disrupting the filament. In this context, we can suspect that MDC#22 would not have enough mass to accrete to form a high-mass star. MDC#10 is in a branch (Fig. A.1) of a thick filament which has started at least low mass star-formation at its other spine-to-branch junction. It is difficult to predict if MDC#10 will be able to grow in mass because it is in a quite isolated area. Towards the filament where MDC#17 and MDC#11 sit, no YSOs are noted. MDC#17, however, being located at a branch to spine junction, appears then as the most favourable starless MDC to form a high-mass star. In addition, on the 350 µm and 8 resolution map (Könyves et al., in prep.), only MDC#11 and MDC#17 are centrally peaked sources, additionnally suggesting they are the most favourable MDCs to form high-mass star.

NGC 6357 and NGC 6334 history
Comparing MDCs for both regions (see Table 7) we can note that they have similar properties except that L bol is about seven times larger for starless MDCs in NGC 6357. Following Ward-Thompson et al. (2002) this higher L bol reflects that the external heating of the starless cores by the local radiation field is higher in NGC 6357 than in NGC 6334. This is expected as NGC 6357 is powered by a the rich OB star cluster Pismis 24 while no such strong radiation source is noted in NGC 6334.
In addition, from his sample of protostellar MDCs, Tigé et al. (2017) estimate a statistical lifetime of 3.5 × 10 5 yr. Such statistical lifetime is estimated from the relative number of a given MDC phase to the OB stars. For NGC 6357 we estimate a total number of 60 O-B3 stars from Russeil et al. (2012Russeil et al. ( , 2017 and Povich et al. (2017). We assume a median age of the O-B3 stars to be 1 × 10 6 yr according to Fang et al. (2012) and Getman et al. (2014). To evaluate the typical lifetime in NGC 6357 we have to estimate the number of massive stars our MDCs can form. Then, following Tigé et al. (2017), and in agreement with Csengeri et al. (2017a), we adopt the same fragmentation level found in Cygnus X protostellar MDCs by Bontemps et al. (2010) to assume that IR-quiet MDCs should host on average two high-mass stars. For high-mass protostellar cores this gives us a statistical lifetime of 3.3 × 10 4 yr corresponding to the free-fall time. This unrealistically small value suggests that the massive star formation has stopped for at least the last Myr and that NGC 6357 will not form any more massive star.
From the Miville-Deschênes et al. (2017) molecular cloud catalogue, based on 12 CO emission, we estimate a total gas mass of 1.6 × 10 5 M (this can be compared with the mass estimation of 2.3 × 10 5 M from Willis et al. 2013 andSchneider et al. 2015) and 2.4 × 10 5 M (which can be compared with the mass estimation of 4 × 10 5 from Cappa et al. 2011) for NGC 6334 and NGC 6357 respectively. The total filament masses are ∼5.4 × 10 4 M and 1.02 × 10 4 M giving that about 25 and 4% of the mass is in the form of filaments in NGC 6334 and NGC 6357, respectively. Considering the total MDC mass, we estimate that about 9 and 14% of the filament mass is in the form of MDCs, for both regions respectively. Comparatively, the estimated total massive core formation efficienty (TCFE) is about 3 and 0.7% for NGC 6334 and NGC 6357, respectively.
The young stellar population also appears very different in the two regions. While the young stellar clusters, identified by Kuhn et al. (2015), are mainly located along the NGC 6334's ridge and hub, in NGC 6357 they are either associated to the clusters Pismis 24 or AH03J1725-34.4. In parallel, Getman et al. (2014 underline an age gradient (between 2.3 and 0.7 Myr and from the south-west to the north-east) of the young stellar cluster along the NGC 6334's filaments while this is not the case in NGC 6357 (clusters and the uniformly distributed stellar population have similar ages between 1.0 and 1.5 Myr), suggesting that the recent star formation proceeded nearly simultaneously across NGC 6357. However, because the Wolf-Rayet phase is expected to occur at a stellar age of 3 Myr (e.g. Sokal et al. 2016) and because OB stars with ages around 4.6 Myr were also reported towards NGC 6357 , we can suspect that starformation in NGC 6357 has been active for at least 5 Myr. The large (radius 15 pc) Hα ring (as defined by Massi et al. 2015) could even have been shaped by a previous event as Russeil et al. (2017) note shock-heated gas towards the filaments on its northeast side. Adopting an expansion velocity of 5 km s −1 would give and age of 3.6 Myr for this ring, which could make it a relic of a previous massive star formation episode.
So we can speculate that previous feedback and the present feedback from O-type stars in Pismis 24 (e.g. Massey et al. 2001 list in Pismis-24 at least two O3 and one WR stars), have stopped the star formation by dispersing its molecular cloud. This is in agreement with the results of Walch et al. (2012) who show that a single O7 star is able to photo-ionise and disperse a 10 4 M molecular cloud in 1-2 Myr. NGC 6334 harbouring fewer and less-massive stars (e.g. Persi & Tapia 2008) the gas removal timescale would then be longer. In addition, NGC 6357 follows qualitatively the evolutionnary picture of the star formation (at kpc scale) with EUV and SN feedback as simulated by Butler et al. (2017). They show that feedback tends to disperse the clustering of the star-formation and to reduce the star formation rate (especially when the mechanical feedback from radiation and supernovae is combined).
On a larger scale, a 100-pc scale feature can be underlined by connecting the young stellar clusters located in NGC 6334 and NGC 6357 (Fang et al. 2012;Kuhn et al. 2014Kuhn et al. , 2015Massi et al. 2015), the ridge in NGC 6334, the filament (at l,b = 353.5 • , +0.66 • ) in NGC 6357, the filament (seen in on extinction map) connecting NGC 6334 and NGC 6357 (Russeil et al. 2010) and the G350.54+0.69 filament . They trace a 100-pc long feature aligned with the Galactic plane at b ∼ 0.67 • (20 pc above the Galactic plane at a distance of 1.75 kpc) which could trace the parental filament of both A134, page 14 of 42 star-forming regions. In this scheme, following Fukui et al. (2018) the formation of NGC 6357 and NGC 6334 could have been triggered by a 100 pc-scale cloud-cloud collision.

Conclusions
In the framework of the Herschel/HOBYS key programme and in the same way as is done for NGC 6334 by Tigé et al. (2017), we performed a study of the massive dense cores in NGC 6357 to better understand how high-mass stars form. We combined the Herschel/HOBYS images to mid-infrared and (sub-)millimeter ground-based data to obtain a complete census of 23 MDCs, among which five are expected to be the most probable progenitors of high-mass stars at 0.1 pc scale. These starless MDCs belong mainly to the edge of the Pismis-24 cavity and the region H II 353.09+0.63 (excited by AH03J1726-34.4). We confirm that, contrarily to NGC 6334, no ridge and hub which can feed the MDCs are observed in NGC 6357. Filaments in NGC 6334 reach higher mass and higher M line than in NGC 6357. In both regions MDCs sit in the most massive filaments underlying their importance for the formation of massive stars in agreement with the empirical model by Motte et al. (2018a) in which MDCs and protostars grow in mass from the surrounding gas. In this picture, because the hottest stars in Pismis-24 have disrupted filaments, the massive star formation in NGC 6357 seems to have stopped and we speculate that little massive star formation will occur.  In addition to extracting filaments with the Vialactea Filamentary Structures Extraction Package (hereafter VFSEP, Schisano et al. 2014 and in prep.) we also performed the extraction using getfilaments a multi-scale, multi-wavelength filament extraction method (Men'shchikov 2013). getfilaments analyses decompositions of original image (here the column density map) across a wide range of spatial scales, the latter being separated by a small amount (a factor of ∼1.05) while the Vialactea Filamentary Structures Extraction Package defines a filament as a two-dimensional elongated region with a relatively higher brightness contrast with respect to its surroundings, and uses a differential method (investigating the eigenvalues of the Hessian matrix of the intensity field) directly related to the contrast. We note that the two methods are in overall agreement and that VFSEP resolved small marginally elongated structures. All of the MDCs are within a VFSEP filament while about half of them do not belong a filament when using getfilaments. In the paper we have used the VFSEP filaments but Schisano et al. (2014) indicate that relatively roundish structures like large and elongated compact clumps, or clusters of compact objects lying on a strong intensity field, might also be detected. To discard such "roundish" structures, we then consider only filaments with length to width ratio larger than 2.