STAR FORMATION AND YOUNG STELLAR CONTENT IN THE W3 GIANT MOLECULAR CLOUD

We aimed to provide the most reliable sample of YSOs in this GMC. For the main classification in our analysis we chose the "revised," updated criteria in Appendix A of Gutermuth et al. (2009). The color and magnitude scheme developed by these authors includes a series of sequential steps (phases) to identify, clean, and classify the candidates. Phase 1: removal of contaminants such as star-forming galaxies with strong polycyclic aromatic hydrocarbon (PAH) emission, active galactic nucleus (AGN), unresolved knots of shock emission, and PAH-emission-contaminated apertures. This step also includes the first separation of YSOs by means of the four IRAC bands. Phase 2: a search for additional YSOs based on 2MASS photometry.¹¹ Phase 3: identification and re-classification of previously identified sources with suitable MIPS 24 μm photometry. When re-classifying photospheric sources into "transition disk" objects (those with significant 24 μm emission; included within the Class II category), we required sources to have been classified as photospheric in both previous phases. To deredden the magnitudes we used the extinction maps and methodology described in Rowles & Froebrich (2009). The visual extinction map was transformed to a median A_H map using the extinction law from Mathis (1990). We changed extinction in the 2MASS bands to extinction in the IRAC channels using the numbers from Flaherty et al. (2007).

The above classification was complemented and cross-checked with (1) the "red source" classification scheme from Robitaille et al. (2008), which should include all Class 0/I and several Class II sources, and (2) the "Stage" phase from Robitaille et al. (2006) (Stage 0/I: $\dot{M}_{\mathrm{env}}/M_{\star }>10^{-6}$ yr⁻¹; Stage II: $\dot{M}_{\mathrm{env}}/M_{\star }<10^{-6}$ yr⁻¹ and M_disk/M_⋆ > 10⁻⁶; Stage III: $\dot{M}_{\mathrm{env}}/M_{\star }<10^{-6}$ yr⁻¹ and M_disk/M_⋆ < 10⁻⁶). We used this last scheme to compare the above observational classification with an alternative method based on intrinsic physical properties (e.g., mass accretion rate and disk mass). This last analysis was carried out by studying the position of each YSO in the color–color diagrams (CCDs) with respect to the limits marking the areas where most of the sources of one particular "stage" are predicted to fall (Robitaille et al. 2006).

Separation and classification of pre-main-sequence (PMS) stars with optically thin disks is a particularly complicated process due to their similarity (in infrared color) with more evolved reddened main-sequence and giant stars (photospheric-dominated). These transition objects are however essential to fully understand the different stages in star formation. In an attempt to estimate the population of sources with weak infrared excess and other PMS stars that may have been missed with the above color classification, we first excluded those objects in our source list already classified as Class 0/I and II. We then used the 2MASS catalog and a process similar to that used in Kerton et al. (2008), who attempted to separate the YSO population guided by a sample of known low-mass T Tauri and intermediate-mass Herbig Ae/Be (HAeBe) stars. In this work (see Section 4), we show that this method can only be applied successfully once the "younger" YSOs have been identified using additional data (e.g., IRAC).

To investigate the possibility of missed candidates from the PMS population we first chose a sample of T Tauri (Kenyon & Hartmann 1995) and HAeBe stars with known distances (Finkenzeller & Mundt 1984; The et al. 1994; Mendigutía et al. 2011 and references therein). All sources were checked with simbad,¹² keeping variable, emission, and PMS stars and rejecting those classified as double/multiple systems, low-mass stars, and brown dwarfs. Infrared photometry was obtained by matching the sample with the 2MASS Point Source Catalog through the gator interface. Infrared photometric systems were converted to the 2MASS system using the transformations from Carpenter (2001). All magnitudes were shifted to a distance of 2 kpc for W3 with the inclusion of interstellar extinction by means of the A_V–distance relation from Indebetouw et al. (2005). Figures 3 and 4 show the CCD and color–magnitude diagram (CMD) for the T Tauri and HAeBe samples shifted to the distance of W3. The CCD shows the T Tauri locus from Meyer et al. (1997) and the main-sequence and giant branch from Koornneef (1983) including interstellar reddening. Reddening vectors for an A_V = 10 have been included for O6 V, M8 V, M2 V, and M6 III stars. The CMD shows solar metallicity isochrones (Marigo et al. 2008; Girardi et al. 2010) at log(t yr⁻¹) = 7, 8, and 9 for the same distance. The dash-dotted line is the reddening vector for an ∼A0 star with A_V = 10 at d =2 kpc, applied using the reddening law and the A_V/A_J conversion from Mathis (1990). The values used for extinction conversions between 2MASS bands are consistent with those from Indebetouw et al. (2005) up to one decimal place taking into account uncertainties, which is a negligible difference compared to the expected uncertainty in the transformation from optical to infrared extinction.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 3 shows T Tauri stars lying mainly above the T Tauri locus. Many are within the reddening band formed by the reddening vectors (Mathis 1990) of an O6 and ∼M2 main-sequence stars (with a tail extending into the HAeBe region, to the right of the reddening band, and following the direction of the T Tauri locus). The wide distribution implies variable amounts of extinctions toward these sources and variable disk emission. A large proportion of T Tauri stars are indistinguishable from main-sequence stars with just interstellar reddening or are consistent with weak-emission T Tauri stars (Meyer et al. 1997). The maximum A_V in the maps from Rowles & Froebrich (2009) is about ∼9.5–10 in the region containing W3 Main/(OH), and ∼7 for those comprising KR 140 and AFGL 333. Thus there will be considerable extinction of objects in the W3 field, and so color identification of this type of T Tauri star without the aid of spectroscopic data will be severely contaminated.

In consequence, we chose a more conservative approach and selected our T Tauri sample by requiring these PMS stars to be reddened enough to lie above the T Tauri locus, with colors satisfying H − K > 0.7 (similar to that used in Kerton et al. 2008 for KR 140). The above limits minimize the contamination from foreground or mildly reddened weak-emission T Tauri stars (undistinguishable from main sequence) and early-type stars. While the color cuts could still allow for non-negligible contamination from late-main-sequence and giant stars with moderate reddening, the magnitude selection criterion below reduces this contamination to mainly that caused by highly extinguished (A_V ⩾ 7) stars of spectral type of ∼A or later (Figure 4).

HAeBe stars lie preferentially to the right of the reddening band. All suitable candidates should therefore be located in this region and satisfy the condition H − K > 0.7. This limit aims to minimize contamination from reddened early-type stars and luminous T Tauri stars. To separate candidates in regions of the CCD populated by both types of PMS (i.e., outside the reddening band) we used information from the CMD (Figure 4) in which T Tauri and HAeBe can be easily separated. All T Tauri stars have K magnitudes ≳ 12. HAeBe stars tend to be brighter, approaching this limit only for "late" stages reaching the main sequence, which we already discard with our imposed limit in the CCD to minimize contamination from reddened early-type stars. The combined color plus magnitude condition minimizes contamination of the T Tauri sample from reddened giant stars. In addition, the color constraint imposed on HAeBe stars, which are mainly localized and "isolated" outside the reddening band of typical stars, already minimizes the contamination from reddened main-sequence and giant stars.

We note that this relatively simple scheme for PMS classification may only be applied after the Class 0/I and Class II populations have been identified using IRAC data, as there is considerable overlap in the CCD and CMD of T Tauri candidates with Spitzer  Class II/I/0 sources (see Section 4).

Our final PMS selection scheme has been summarized in Table 1. The color and magnitude selection criteria were applied to all sources in our initial IRAC list satisfying (1) the cleaning/reliability conditions in the Spitzer channels; (2) matched to a 2MASS source with quality flag better than "D" in all 2MASS bands; and (3) classified as (mainly) photospheric or without a successful classification using the YSO scheme from Gutermuth et al. (2009) (i.e., no Class 0/I, Class II, or contaminant).

A search for additional PMS was also carried out by extending our analysis to 2MASS sources in the area covered by the Spitzer survey but without a suitable IRAC counterpart (i.e., satisfying our initial cleaning and reliability conditions) in the short-wavelength Spitzer channels. The lack of detection in Spitzer  minimizes the possibility of confusion with actual embedded protostars, which should have been previously identified with the color/magnitude criteria from Gutermuth et al. (2009).

The YSO search using the color and magnitude classification from Gutermuth et al. (2009) yielded a total of 616, 706 (two of which were also observed in neighboring Astronomical Observation Requests (AORs)), and 246 YSOs in the regions surveyed in all four IRAC channels in each individual AOR: W3 Main/(OH), KR 140/KR 140-N, and AFGL 333 Spitzer regions, respectively (Figure 5; Table 6). The full list of YSOs (Table 2), 2MASS-based PMS candidates (Table 3), and galaxy candidates (Table 4) appear in the electronic version of this article.

Zoom In Zoom Out Reset image size

For the purpose of the present analysis we chose reliability over completeness, and so this sample requires S/N ⩾5 for any IRAC/MIPS photometry used in any particular method. When using 2MASS photometry in phase 2 of the color/magnitude classification scheme, those sources with the closest distance to our IRAC sources (within 2'') were chosen as suitable counterparts as long as they had a quality flag "A," "B," "C," or "D" in H and K bands, and at least valid photometry ("D") in the J band (if available).

In order to remove as many artifacts and false detections as possible, we first bandmerged the IRAC channels using channel pairs. A candidate source could still be included in the initial list without suitable photometry in all four channels if it either appeared in channels 1 and 2, or channels 3 and 4. This was done as the first step to minimize the need for visual inspection of samples containing tens of thousands of sources, while at the same time attempting to minimize the loss of relatively "cold" sources without a suitable detection at shorter wavelengths. In addition, we also imposed our own "internal" cleaning conditions based on the properties provided by APEX and SExtractor for those sources already published as reliable detections in the catalog of Ruch et al. (2007) (e.g., successful PRF fitting, χ² of fitting, ellipticity, and successful deblending). We note, however, that these internal cleaning parameters are physically meaningless and were only used for "relative" classification of sources within a particular sample, with the only purpose being to reject as many artifacts (e.g., PSF residuals and spikes around bright sources) and false detections as possible. Despite this procedure, these conditions were still conservative, and visual inspection and manual rejection was still required and performed in the last stages of the catalog production process. All sources successfully classified as YSO candidates and satisfying all the cleaning conditions and bandmerging requirements have an entry =1 in the "Flag" column in the source catalog (Table 2). This defines the "reliable" subset of the final list (Figure 5).

In an attempt to improve the "completeness" of the sample we also used the detections in IRAC channels 1 and 2 (with the best sensitivity) as a base for a new source list. Photometry was performed on fixed centroids at longer wavelengths (including MIPS), and all the cleaning/selection conditions and the Gutermuth classification scheme were again applied to each source. Those additional detections satisfying all the catalog requirements were included in the final catalog with a flag entry =0. We note that in the scheme of Gutermuth et al. (2009), MIPS photometry was used mainly to re-classify as YSOs those sources initially rejected as galaxies, AGNs, or fake excess sources in the first steps of the scheme.

Photometry derived from the MIPS 24 μm maps was generally less reliable. We therefore produced a Catalog 2 (Figure 6) based on the same procedure as Catalog 1, but allowing for re-classification of IRAC/2MASS sources only if there was a successful MIPS counterpart in our original (independently obtained) MIPS source list (i.e., the standard fixed-centroid photometry is not used in MIPS phase 3). Catalog 2 is therefore more conservative, because although the SExtractor extraction was visually observed to detect all significant sources, a large fraction of these detections did not satisfy the cleaning conditions after performing APEX PRF fitting on the MIPS mosaics, on account of the variable and complicated background at longer wavelengths. Both catalogs (Catalog 1 and Catalog 2) yield very similar source lists for Class 0/I and Class II sources, differing mainly on the number of Class 0/I* (highly embedded YSOs) and Class II* transition objects. Defined in this work as the (*) population, both the highly embedded and transition objects rely on MIPS photometry for identification and classification. Final photometry for the Spitzer-baser YSO sample has been included in Table 5. A summary of the number of candidates found of each Class in each field is given in Table 6. As expected, highly embedded and transition objects are particularly abundant in Catalog 1 due to the use of MIPS photometry based on IRAC centroids, with Class 0/I* sources forming up to ∼65% of the population in W3 Main/(OH) (∼5.5% Class II*). In Catalog 2 these types of objects constitute less than ∼3.5% of the YSOs in each field.

Zoom In Zoom Out Reset image size

We created a sample resulting from bandmerging just channels 3 and 4 and ran it through the cleaning and classification procedures described above. This experiment produced no new sources. This shows that no significant sample of "cold" sources (without detections in IRAC channels 1 and 2) should have been missed by the initial bandmerging-by-pairs procedure (within the limitations of this technique). We therefore conclude that the majority of sources potentially missed in our (flag =1) analysis would have come from the samples in IRAC channel 1/channel 2 that do not have counterparts at longer wavelengths due to the diffuse emission and sensitivity loss (although we note that Catalogs 1 and 2 for flag =1 are very similar; Table 6).

Unless mentioned otherwise, in the following sections we will use Catalog 1 (all flags) as the primary sample in our analysis. As we explain in Section 5, this source list is expected to be a more reliable indicator of the YSO properties. We find that the use (or omission) of the (*) population is mostly relevant in those highly populated regions with high extinction, mainly IC 1795 and KR 140. We cross-correlated our final YSO list with the list of infrared sources associated with the cluster IC 1795 presented in Roccatagliata et al. (2011), and find 76 YSOs in our sample which are consistent with being cluster members. Of these, 30 YSOs belong to the (*) population in Catalog 1 (not classified as YSOs in Catalog 2), and yet all are also classified as YSOs (Class II) in Roccatagliata et al. (2011). When applying the color classification from Megeath et al. (2004) we find that only 4 out of the 76 sources would not have been classified as YSOs, which supports our decision and the need to keep the (*) population in our analysis. Results based on Catalog 2 are only mentioned briefly when required.

We finally note that while the flag =0 and IRAC short-wavelength-based catalogs intend to improve the completeness of the final sample (and each source was visually inspected in channel 1), these detections are still tentative and should be treated with caution.

We compared Catalog 1 and Catalog 2 with the detections from Ruch et al. (2007) for W3 Main/(OH), the region analyzed in their study. These authors detected a total of 295 sources in the four IRAC channels, 21 (∼7%) of which were classified as Class I sources, and 94 (∼32%) as Class II. All 295 sources were in our initial catalog resulting from the SExtractor source detection process. Our analysis identifies a similar number of Class I and Class II candidates in this sample, with a total of 117 sources classified as YSOs in Catalog 1 (93 in Catalog 2). Of the remaining sources in their list not classified as YSOs in Catalog 1, 130 are stars, 0 galaxies, 3 shock/knots of emission or sources with PAH-contaminated apertures, and 45 do not satisfy the cleaning and reliability conditions. Sources in their sample not classified as YSOs in Catalog 2 consist of 149 stars, 0 galaxies, 3 shock/contaminated aperture sources, and 50 sources not satisfying cleaning/reliability conditions. In all cases, the percentage of Class II sources is ∼3 times that of Class 0/I sources. Table 7 shows the results after applying our cleaning conditions and the Gutermuth scheme.

Table 6 includes the number of PMS stars found based on 2MASS color and magnitude information, with counterparts in the Spitzer images. We found 51 more T Tauri stars and 2 HAeBe stars when searching for 2MASS sources (from the 2MASS PSC) without suitable IRAC counterparts (i.e., IRAC sources satisfying our initial cleaning conditions), but located in the common (four channels) areas surveyed by Spitzer. Many of these sources are located in the bright, diffuse region surrounding W3 Main, (OH), and AFGL 333 in the IRAC images, which did not allow for proper identification and extraction of candidate sources due to confusion. The imposed lower limit for the K magnitude of our HAeBe stars is within the 2MASS 10σ completeness limit for this band (14.3 mag). This sample is also complete in the H band (15.0 mag) and the J band (15.9 mag). The faintest 2MASS magnitudes for the T Tauri sample lie beyond the 2MASS completeness limits by ∼1 mag in K and H bands, and ∼2 mag in the J band, and therefore our list will be incomplete at the faint end of the population.

The use of several steps of cleaning and reliability thresholds for our final sample, combined with different sensitivity limits in different regions for a given field, complicate the determination of a reliable completeness limit for our catalog. In order to provide an estimate of the magnitude at which a catalog is 100% complete, which due to our cleaning and selection steps differs from a fainter limit at which some sources can be detected, we examined the number of detections as a function of magnitude for each field at each IRAC wavelength. Figure 7 shows histograms for W3 Main and W3 (OH) for all YSOs in Catalog 1, the most complete sample derived in this work. Our estimates for 100% completeness, given by the "turnover" points in the YSO distributions for each subregion of W3, are included in Table 8. The effects of strong, large-scale emission in dense regions with high stellar activity is evident from the limits derived for the HDL.

Zoom In Zoom Out Reset image size

Our choice of reliability over completeness excludes a large population of faint sources from our final list, and therefore the completeness limits for the different regions reveal a particularly conservative population for IRAC channels 1 and 2. The flag =0 sample (using the short-wavelength channels as a template for photometry at longer wavelengths) improves completeness at these channels significantly, although we note that our estimated completeness limits are still conservative compared to samples derived from similar Spitzer studies of W3 (Polychroni et al. 2010). Clearly, the completeness magnitudes quoted for our work are upper (faint) limits, especially for the long-wavelength channels. The use of flag =1 data or the sample from Catalog 2 would be more conservative (and therefore less complete). The latter candidate source list, while more reliable than Catalog 1, is expected to suffer a more severe loss of highly embedded sources. This is due particularly to the rejection of poorly fitted sources (failed PRF fitting) with APEX, and our strict requirement of suitable fitting and photometry in the MIPS band (with severe sensitivity loss and confusion due to strong diffuse emission). We expect to compensate for this issue with the Herschel data currently being analyzed, which will be able to constrain the dust emission much more accurately for those embedded (younger) sources.

With regard to Class types, Class II sources are generally weaker at the longer IRAC wavelengths than more embedded Class 0/I objects, and are therefore more likely to be missed because of lower sensitivity and bright PAH emission (e.g., Chavarría et al. 2008).

Despite our attempt to compensate for saturation in the IRAC sample by using the short exposure mosaics, some sources will still be missed in the brightest and most active regions such as W3 Main and W3 (OH) due to confusion and location within the bright infrared nebulosity. The catalogs would also exclude clearly extended sources and stellar groups.

Due to our attempt to provide a reliable sample for our PMS sources, we again note that weak-lined and low-extinction T Tauri and HAeBe stars have been excluded from our catalog in order to avoid major contamination from reddened main-sequence stars. We cross-checked our final list of candidate T Tauri and HAeBe stars with the simbad database, and found no matches with the exception of a couple of sources, classified in the database as "infrared sources" or "star in cluster." Ultimately, spectroscopic surveys of stellar candidates are the most reliable method to find and confirm PMS.

The Spitzer-based sample was subjected to strict cleaning conditions and cross-checked with other classification schemes to ensure the highest consistency and reliability of our sample of YSO candidates. However, it is likely that some contamination from extraneous objects, such as galaxies/AGNs, planetary nebulae (PNe), and asymptotic giant branch (AGB) stars, is present in the final sample.

A measure of the contamination by PNe and AGB stars was obtained by applying the conditions from Robitaille et al. (2008) to the reliable photometry derived from the previous YSO classification. For successful classification as a PN, detections are needed to satisfy at least two of the four color conditions to ensure reliability in their position in the CCDs. While the selection scheme still allows for mutual contamination of YSO/AGB stars in both samples, this technique can still provide a useful measure of the general contamination of our YSO sample. The PN contamination using this method is predicted to be ⩽1.5% for W3 Main/(OH), and ⩽1% and ⩽2% for KR 140 and AFGL 333, respectively. The contamination from AGB stars is expected to be <3.0% and <0.5% for W3 Main/(OH) and AFGL 333. No candidate AGB stars are found in KR 140 using this scheme.

In their analysis, Gutermuth et al. (2009) adapted their classification to work with the larger distances present in their survey of up to ∼1 kpc. The use of this classification at the distance of W3 (∼2 kpc) is expected to shift the proportion of classified YSOs toward the brightest sources, because of the loss of dimmer sources rejected through the magnitude limits imposed on the sample and a possible contamination of the galaxy sample with YSOs. To check this effect, we ran the codes on the YSO sample in W5 from the work of Koenig et al. (2008), applying the new classification conditions to the photometry provided by these authors. W5 belongs to a massive complex (neighboring W3 and W4) and is considered to be at the same distance as these GMCs. These authors found a significant proportion of YSOs being misclassified as non YSOs, which was evident, for instance, in the "clustered" properties of the "galaxies."

Although the code used in the present work is a "revised" version of that used by Koenig et al. (2008), we successfully classified ∼95.5% of their sources as YSOs of the same type (excluding the very few sources not satisfying the requirement of magnitude error <0.2 mag). About two thirds of the 4.5% sources where we disagreed with the classification from Koenig et al. (2008) were classified as AGNs, and 0% as PAH galaxies. Increasing the magnitude limits for AGN classification by 0.5 mag to account for the larger distance of W3 (e.g., Megeath et al. 2009), we find no differences in the percentage of sources classified as AGNs. Although we cannot perform this test on their rejected (non-YSO) sample (as that data was not included in their publication), we expect this code to detect and recover successfully the main YSO population in W3, without the need for major additional modifications.

This is in agreement with the following analysis of the distribution of those sources classified as galaxies or AGNs. Using the "distance to nearest neighbor" technique (Clark & Evans 1954) we measured the mean observed distance between galaxy candidates relative to the mean distance that would be expected for a random distribution for a population with the same characteristics. The ratio between these two quantities (R) would be equal to one for a perfectly random distribution. We obtain R =0.94, 0.98, and 0.88, with significance levels of ∼25%, ∼37%, and ∼5% for W3 Main/(OH), KR 140, and AFGL 333, respectively. This supports the random spatial distribution of our Galaxy candidates and low YSO contamination based on the lack of significant clustering. Coordinates for the list of galaxy candidates are included in Table 4.

In Section 3, we established a scheme to separate the populations of T Tauri and HAeBe stars based on magnitude and color information. HAeBe stars occupy a distinct region in the CCD, but K magnitude data is still essential to separate the two samples in the region near the T Tauri locus populated by both types.

We carried out T-test and F-test analyses comparing the color and magnitude distributions of various Classes. Results are reported in Table 9. Statistically, and as in previous cases, the Class 0/I, Class II, and T Tauri samples are indistinguishable in the 2MASS CMD, and therefore selection of low-mass PMS stars is not possible in the infrared without prior knowledge of the protostar population. In color, the Class 0/I sample is intrinsically redder, with the main population of Class II sources lying more intermediate between Class 0/I (and closer) to the T Tauri sample (Figures 9, 12, and 10). ∼90% of Class 0/I sources lie between 0.7 ⩽ [H − K] < 2.5, compared to Class II sources, located within 0.4 ⩽ [H − K] < 1.7. While both have similar maximum [H − K] value of ∼3.4, ∼80% and ∼50% of Class 0/I sources lie above [H − K] =1.0 and 1.5, respectively, compared to ∼30% and ∼9% for the Class II population.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We find a completely different scenario when focusing on the intermediate-mass HAeBe population. As shown in Figure 4, Herbig stars are not only redder, but also brighter than T Tauri stars, with both populations clearly separated in the CMD. Class 0/I candidates are not consistent with the colors or magnitudes of HAeBe stars (Class 0/I being redder and dimmer). Class II sources are consistent in color with Herbig stars (i.e., intermediate between T Tauri and Class 0/I), but are again dimmer than typical HAeBe stars (Figure 13). While a spectroscopic analysis is still required to confirm our HAeBe candidates as such, we find that, contrary to T Tauri stars, Herbig stars may still be selected without prior knowledge of the embedded population.

Using the 2MASS data we now analyze the Stage I and Stage II populations (as defined in Robitaille et al. 2006). F-test significance levels between Stage I–Stage II and the T Tauri population show both groups are statistically consistent with having been drawn from the same parent population in K magnitude as the T Tauri sample (sig. levels: 0.41 and 0.58, respectively), with Stage II also consistent in color (sig. level: 0.39) with the PMS population (dominating in the reddening band of typical main-sequence and giant stars). Stage I sources show a larger scatter in the CCD (e.g., Figures 11 and 14), and significance levels obtained from Student's t-test and F-test between the two populations agree with both having different parent distributions based on color information (Table 10).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The populations in Stage I and Stage II, when treated as a whole, are indistinguishable with respect to observed K magnitude (e.g., Figures 15, 16, and 17). However, while Stage I sources are found in the color region occupied by Stage II, members of the former population consistently reach redder colors: ∼90% of Stage I sources have [H − K] >0.6, ∼85% have [H − K] >0.7, ∼55% have [H − K] >1.0, and ∼25% have [H − K] >1.5 (maximum [H − K] =  ∼ 3.5). The corresponding statistics for Stage II are ∼65%, ∼40%, 10%, and 2% (maximum [H − K] =2.2).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We find no clear boundary in the CCD or CMD separating sources with intrinsic, physically different characteristics, which could easily be explained by different inclination angles. About 90% of Stage 0/I sources lie in the range 0.5 ⩽ [H − K] <2.0, ∼90% of Stage II between 0.4 ⩽ [H − K] <1.2, leaving the region in [H − K] ⩾1.2 as the area dominated by (younger) sources with $\dot{M}_{\mathrm{env}}/M_{\star }>10^{-6}$ yr⁻¹ (e.g., Figure 17).

Zoom In Zoom Out Reset image size

The cluster IC 1795 is located at the center of an eroded shell-like structure containing the W3 Main and W3 (OH) complexes. Both show the most active massive star formation, including deeply embedded clusters, H ii regions, and a trapezium-like system rivaling that in the Orion nebula (e.g., Megeath et al. 2005).

Figure 19 shows the W3 Main/(OH) field with the identified groups, the contours representing the surface density distributions for YSOs calculated on a grid identical to that of the Spitzer images, and including highly embedded and transition stage candidates (Class 0/I* and Class II*: (*) sample). This figure also includes "age" contours, obtained from the ratio of the surface density maps: [Class II + Class II*]/[Class 0/I + Class 0/I*].

Zoom In Zoom Out Reset image size

Figure 20 shows the groups found for the same N_YSO and break length as Figure 19, but excluding the less reliable (*) population. For convenience these will be specifically referred to as "magenta" groups, to distinguish them from the "red" groups in Figure 19. Reduction of the number of candidate YSOs yielded smaller and fewer groups, as expected. In addition, the omission of highly embedded sources, combined with the confusion in the regions of strong IR emission, highly affected the number of YSOs detected in the innermost and most active regions, especially around W3 Main. We find two magenta groups within red Group 1, coincident with IC 1795. The oldest (magenta Group 2; Figure 20) coincides with the oldest part of IC 1795. Magenta Group 0 coincides with a high extinction region and contains numerous H ii regions and known clusters, including the well-known maser sources W3 (OH) and W3 (H₂O). Although with N_m < 25, the characteristics of these two groups are consistent with those of mid-rich triggered groups in Figure 19 (see below) linked to massive star formation, with surface densities greater than 70 M_☉ pc⁻², A_V-mean > 5.0, and mean inter-YSO separations D < 0.25 pc.

Zoom In Zoom Out Reset image size

In Figures 21 and 22, we show the groups identified for each particular Class (with and without the (*) population) with the relaxed requirement of N_YSO = 5. The identified groups more closely trace particular structures observed in extinction (e.g., filaments), as well as smaller separate concentrations of Class 0/I and Class II candidates.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We suggest that, overall, clustered formation in compact clumps/cores can be distinguished from the star formation process associated with young groups that have low surface density and extinction and that are found in triggered regions (e.g., Groups 6 and 7; Figure 19). Figure 23 shows the inter-YSO separations as a function of SFE. Groups in "induced" regions with the largest separations have been less efficient in forming young stars, which could be due to formation in a turbulent environment. This could be indicative of the "distributed" star formation mode taking over in those regions that lack the conditions, like high column density and low turbulence, required to form massive stars and their (richer) parent clusters. Indeed, a simbad search reveals no H ii regions or clusters associated with any of these triggered groups, which also lack localized and significant radio continuum emission in the Canadian Galactic Plane Survey (CGPS; Taylor et al. 2003) maps. Rich groups (25 ⩽ N_m < 50) such as Group 0 and Group 5 have average inter-YSO separations <0.25 pc and the highest surface densities, and both contain within their perimeters a large proportion of the major massive star activity of the W3 GMC. This supports the classical view that one can have clustered star formation caused by triggering when at early stages high surface densities without disruptive turbulence are present, to ensure that a gravitationally (unstable) bound system is formed.

Zoom In Zoom Out Reset image size

The privileged location of IC 1795 (Group 1; Figure 19), in the most central and dense parts of the HDL, was likely key in the formation of its rich population. Although it depends on the pre-existence of the HDL, its location, morphology, and population characteristics are compatible with isolated/quiescent formation, and not necessarily the product of a triggering event as suggested in previous studies. A similar (quiescent) origin is also suggested for some individual isolated groups in the outer boundaries of the field, characterized by a relatively poor and yet closely spaced (contrary to the poor groups associated with triggered regions discussed above) old population in a relatively compact configuration (e.g., Group 2 in Figure 19). Group 1 is the next "oldest" system after Group 2, and it is the most massive and richest group, with the smallest separation distances down to 0.02 pc.

W3 Main and W3 (OH) (Figure 1) show the highest extinction and both are likely to have (at least part) of their stellar population induced by IC 1795, in agreement with the conclusions from Oey et al. (2005) and Polychroni et al. (2010). A triggered scenario is supported by (1) an elongated "older" Class II dominated region within Group 1 extending in the directions of W3 Main and W3 (OH) (see Figure 19); (2) the cavity and shell-like structure around IC 1795 hosting both systems; and (3) a tendency of Class II candidates to be located toward the most central regions, with younger groups of YSOs following the shell around Group 1 (in increasing age: Groups 4, 7, 0, 5, 3, 6; Figure 19). Using the ratio of Class II/Class 0/I as a relative measurement of age, Group 6, the oldest after Group 2 and Group 1, happens to be also the nearest to IC 1795, while younger groups have larger distances.

Secondary bursts of star formation, as well as a non-negligible star formation duration, are needed to reconcile the characteristics of the YSO population in IC 1795 with the age estimated by Oey et al. (2005, 3–5 Myr) using optical photometry and spectral analysis. We find a highly distributed population of Class II and Class 0/I sources in the vicinity of IC 1795. Star formation is estimated to occur in 1–2 dynamical crossing times (Elmegreen 2000), but larger age spreads are possible if multiple events occur within a certain system. Figures 21 and 22 show the groups obtained when separating the YSOs according to Class with a minimum group membership requirement of N_YSO = 5. The "class subgroups" observed within Group 1 and the triggered population could be indicative of such additional star formation events. This combined activity likely reinforced the effects of the central cluster when forming the cavity and dense surrounding shell hosting the most massive stars in W3.

Contrary to the suggestion of Oey et al. (2005), some of the stellar activity in W3 Main might also have originally initiated in quiescent mode. Figure 24 shows an "older" (low- and intermediate-mass) PMS population toward the outer edge of this region, shown by the magenta/red contours. On the other hand, there are young groups at the inner edge of W3 Main, closer to IC 1795. By calibrating the Class II/Class 0/I ratio of Group 1 to the age of IC 1795, we obtain an average age for these young groups of ∼1.5–2.5 Myr. In this calculation we assumed the onset of star formation occurred in a single event, and so a larger proportion of Class II sources implies a more evolved population. We ignored effects such as secondary bursts of star formation in the region, and assumed that for groups with similar initial stellar characteristics there is a direct relation between the Class ratio and the actual age of the system. A non-negligible period of star formation and internal triggering would mean that it would be more appropriate to consider our age estimates as lower limits. This age estimate for the young groups is however comparable to the estimated age for the oldest (diffuse) H ii regions and the PMS population known to dominate in this region (∼10⁶ yr; Feigelson & Townsley 2008). Therefore, triggering (either indirectly by IC 1795, and/or by this pre-existing PMS population) could have been responsible for the observed young proto-OB population in the central regions of W3 Main, a secondary burst of massive star formation in a region with an already highly enhanced surface density. Such a scenario would nicely link IC 1795 to one of the possible models suggested by Feigelson & Townsley (2008) to explain the origin of W3 Main (option 4 in their list), and could explain the anomalous age distribution (a central young cluster surrounded by an older population) described by these authors.

Zoom In Zoom Out Reset image size

W3 Main and W3 (OH) dominate both in massive stars and in star formation activity. The other HDL field, containing AFGL 333, is characterized by a young stellar population in its most part associated with high extinction regions and filamentary structures that have remained undetected in the 2MASS-based extinction maps. There are also clear similarities and some differences relative to W3 Main/(OH).

Figures 25 and 26 show the identified YSO groups (with and without the (*) population) for the AFGL 333 field, including YSO surface density contours and age contours. Figures 27 and 28 show the groups identified for N_YSO = 5 and different Classes (each with their corresponding D_break). The oldest and youngest groups are clearly separated, with the youngest (brown ellipses) localized in the central regions and the oldest (blue ellipses) toward the outer parts of AFGL 333 (northern parts and triggered regions neighboring W4).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Group 4 (Figure 25) is the oldest and most massive group in the field, with the largest area, surface density, and highest extinction of ∼A_V-peak ∼ 7.5). From our analysis we conclude that its population formed first, clearing a region and eroding a cavity-like structure much like IC 1795 (sizes ∼ 6–7 pc) but in YSOs less populated. Within this cavity, the YSO population shows a more compact configuration in the eastern side of the group, and a more widely separated population toward the western side, where some high extinction filamentary structures are identifiable. There is also a population of PMS candidates toward the edge of this cavity, on the western side of Group 4 (opposite W4). Just as with IC 1795, the properties of this group (its location in the middle of the HDL and the geometrically "ordered" population, already eroding the surrounding material) are compatible with it being produced by quiescent formation, rather than having a triggered origin by the Perseus superbubble.

A significant difference relative to W3 Main/(OH) is the abundance of filaments with and without stellar activity. Groups in this field are all relatively poor, with only two groups (red Groups 1 and 4) with N > 25 members (Figure 25) and with the YSO population mainly localized to these filamentary structures. The morphology of such filaments, seen well in Spitzer channel 4 (Figure 2), suggests a turbulent origin, a suggestion supported by their remarkable similarity and continuity with nearby (non-extinction) structures. A good example is Group 6 (Figure 25), associated with a filamentary structure within the cavity caused by the stellar population in Group 4, and containing a relatively old YSO population (but still younger than the latter). Its associated population, while slightly offset from that filament either due to displacement since formation or clearing of the immediate surroundings, clearly traces the overall shape of the high extinction, indicating a birth association. While the filament in Group 6 is easily traced, only dispersed "remnants" of high extinction structures are visible at the western side of Group 4. This supports the "older" evolutionary stage for the latter, where the apparently dissociated population may be due to the dispersal of their common parental filaments.

A triggered origin is clearly identified for Groups 7 and 3 (N_m < 25), both in the outer boundary of the HDL and associated within structures carved by the activity in W4. The pillar associated with Group 7 coincides with a known cluster (IRAS 02252+6120; Bica et al. 2003), containing both a Class II population, located toward the outer edges of the pillar, and Class 0/I sources, mainly in the innermost regions. Such a configuration has previously been observed in other pillar-like structures (e.g., Choudhury et al. 2010) and suggested to be the result of radiative-driven implosion (RDI) triggered star formation, in which an ionization/shock front driven by an expanding H ii region causes a neighboring overdensity to collapse, triggering the formation of stars.

Group 1 is the youngest and the richest group in Figure 25. It is located in the innermost regions of AFGL 333 and is associated with the most dense and prominent filament in this field. If filaments are formed (or induced to collapse) by turbulence and compression by nearby star activity, then Group 1 might have been induced by the activity at opposite sides associated with IRAS 02245+6115 and IRAS 02244+6117, bottom and top crosses in Figure 25, respectively. The latter contains bright infrared stars (BIRS; Elmegreen 1980), and the former hosts a known cluster and massive star activity (Bica et al. 2003). These form the brightest regions in the mid-infrared in AFGL 333 and both contain a population of PMS in their outermost parts. Thus, two active older regions sandwich the central young stellar population of Group 1.

Group 5 (Figure 25), while associated with the same high extinction structure as Group 1, is however relatively member-poor. Its location away from the influence of the main (infrared-bright) star-forming activity in AFGL 333 may explain the low membership, in contrast with the richness of Group 1.

High extinction structures border Group 0 as well (especially noticeable in IRAC channel 4; Figure 2). The associated filaments are located within an evacuated region, much like those in Group 6. However, as mentioned above, triggering in such regions of low surface density will likely result in isolated star formation (instead of clustered).

The results from the above analysis remain unchanged when the (*) population is excluded, except for the omission of red Groups 2 and 5 (Figure 26).

We note that the features seen in extinction in the mid-infrared are better tracers of high column density material than the 2MASS-based extinction map, the latter missing some major high extinction areas (e.g., Group 1, Group 7; Figure 25). Thus, the mass (and surface density) estimates for associated groups are lower limits. Indeed, Group 1 has the largest number of associated YSOs, and it is expected to have a large surface density comparable to or greater than groups of similar membership, >80 M_☉ pc⁻². The extinction structure associated with this group is the brightest in the AFGL 333 field in the 850 μm SCUBA map, which is an excellent tracer of column density. We find a similar situation in the KR 140 field (below). For small structures and filaments prominent in the submillimeter maps, including KR 140-N, north of the KR 140 H ii region near the center of the field, information is lost in the 2MASS-based extinction maps. Our Herschel analysis (A. Rivera-Ingraham et al. 2011, in preparation) will be able to provide accurate masses and properties for each of these structures based on temperature-calibrated dust emission.

It has been argued that the presence of massive star-forming sites along the interface between W3 and W4 is evidence that the main activity in the HDL is triggered by the expansion of the W4 H ii region (e.g., Carpenter et al. 2000 and references therein). This is supported by the estimated ages of the structures (e.g., Oey et al. 2005). We agree that the HDL structure itself was created by W4, with conditions (e.g., turbulence, surface density) favorable for star/cluster formation. However, whether the formation and ultimate collapse of clumps and cores in W3 were directly triggered is less clear. Based on the above evidence it seems also plausible to us that formation of the HDL was followed in sequence by a more quiescent evolution governed by local conditions.

Local triggering does play a major role in enhancing (massive) star/cluster formation: (1) internal triggering within evacuated regions (inner shells) generates secondary bursts of star formation and a distributed population either by forming or collapsing pre-existing small overdensities (clumps and filaments; IC 1795, AFGL-Group 4, AFGL-Group 0); (2) compression of high-density regions generates major bursts of star formation, including massive stars and highly embedded clusters (e.g., inner part of W3-Main, W3 (OH), AFGL-Group 1, AFGL-Group 7).

Overall, the star formation activity and processes in the AFGL 333 and W3 Main/(OH) fields operate similarly, albeit less vigorously in the former. Environmental physical differences between the two regions, such as column density distribution and kinematics, will be investigated in upcoming papers. The preponderance of filaments and the characteristics of these and other star forming and starless structures in the HDL will be the subject of our Herschel paper currently in preparation.