In order to characterize the progress of star formation throughout W5 it is important to establish the evolutionary status of stars within the region. The presence of massive O stars (Hillwig et al. 2006) and significant molecular material in W5 (Lada et al. 1978), suggest that the stars associated with the region will be of relatively young age (<10⁷ yr). Recently formed stars exhibit infrared excess emission in their spectra above that produced by the stellar photosphere. This excess emission arises from heated dust, either within the circumstellar material close to the young star, perhaps falling onto it as a part of protostellar collapse, or left behind after the end of star formation. This material gradually disappears with time, and hence the excess emission evolves as a function of stellar age. We make use of this by measuring (see, e.g., Lada 1987; Stahler & Palla 2005): the value of which decreases with the progression of the star’s evolutionary state, whether through its own aging or the effects of its environment.

In the case of most surveys of star-forming regions, including W5, infrared colors serve as proxies for directly measuring . Schemes developed in Whitney et al. (2003), Allen et al. (2004), Megeath et al. (2004), and Muzerolle et al. (2004) and tested by (for example) Hartmann et al. (2005) have demonstrated the power of this technique for classifying young stellar objects using Spitzer IRAC and MIPS photometry. The classification scheme we use is described fully in Gutermuth et al. (2008a). We use an iterative method combining IRAC and MIPS photometry with an extinction map generated from 2MASS near-IR photometry (Skrutskie et al. 2006). The scheme classifies the stars and filters out extragalactic contamination.

We generate an extinction map from 2MASS colors of stars detected across the entire W5 field using the method described in Gutermuth et al. (2005), which itself is based on the NICE and NICER algorithms of Lada et al. (1994) and Lombardi & Alves (2001). The map has an angular resolution of ∼ and is sensitive up to . However, the map is limited by the sensitivity of the 2MASS survey. As a result, values in the map above ∼3 may be underestimates of the true extinguishing column through the entire cloud, since objects behind are too faint to be detected. This is only the case for <1% of pixels in the map, so we ignore this effect for our analysis. With values in hand for all sources, we deredden their IRAC magnitudes using the IR extinction law presented in Flaherty et al. (2007), assuming each source is seen behind the full extinguishing column in each case.

We first remove star-forming (PAH) galaxies and weak-line AGNs via a series of cuts in the four-band IRAC color-color diagrams after the procedure developed by Gutermuth et al. (2008a). Because W5 is ∼4 times more distant than the nearby regions studied by Gutermuth et al., this filter removes many apparent young stellar objects (YSOs). In the Appendix we discuss how we characterize and account for this effect in our subsequent analysis. Next we filter out unresolved shocked blobs of PAH emission by cutting objects with a large 4.5 μm excess, i.e., very red color (Smith et al. 2006). We categorize the remaining—presumably stellar—sample primarily relying largely on the color to discriminate among SED classes. In Figure 5 we show IRAC color-color diagrams for sources in W5, marking the location of protostars (Class I/0, red dots), stars with disks (Class II, green dots), and stars exhibiting only photospheric colors (Class III, black dots) as identified using our scheme.

Figure 5. Left: vs. ; right: vs. IRAC color-color diagrams used for identifying candidate protostars. Black dots: Class III; green: Class II; red: Class I; yellow: transition disk candidates. AGN and PAH galaxy candidates are not included here.

We also use 2MASS and photometry, combined with IRAC 3.6 and 4.5 μm data to classify objects lacking either an IRAC or detection. To make sure the 2MASS sources have reliable photometry we require a magnitude error ⩽0.1 in both and bands. We first deredden the photometry in these four bands and then identify IR excess sources as those having red and colors. The results are shown in Figure 6 (left).

Figure 6. Left: Dereddened 2MASS vs. color-magnitude diagram. The dashed red box shows the location of AGNs detected by the IRAC shallow survey (Eisenhardt et al. 2004). Right: IRAC and MIPS vs. color-color diagram. Black dots: Class III; green; Class II; red: Class I; yellow: transition disk candidates. Suspected AGN and PAH galaxies are not included here.

MIPS 24 μm photometry provides us with additional classification information. Figure 6 (right) shows the color-color diagram combining MIPS and IRAC photometry. Stellar sources not classified as either Class I or Class II by their IRAC or near-IR colors may still have red colors ( ) and thus be candidate “cold disk” or “transition disk” objects—in other words, Class II objects with significant clearing of their inner disk region (see, for example, Muzerolle et al. 2004; Lada et al. 2006; Cieza et al. 2007; Najita et al. 2007; Flaherty & Muzerolle 2008; Brown et al. 2008). These are marked with yellow dots in Figures 5 and 6. MIPS photometry is also used as a check on the AGN/galaxy/shocked-blob filtering and Class I classification. AGN candidates with bright MIPS detection can be reclassified as protostars, and Class I objects with insufficiently red or colors are demoted to Class II.

Finally, we also visually inspected the SEDs of all objects in Class I and Class II (a total of 2064 objects), where available making use of 2MASS and/or 24 μm photometry. In most cases only IRAC four-band data were present. We give the full list of objects, Class I, Class II, Class III, and transition disk candidates, in Table 2. We list source coordinates and photometry in near-IR (from 2MASS) and Spitzer bands and give a column denoting infrared SED source class and a flag column for objects with unusual SEDs. Objects are sorted and indexed by ascending right-ascension order. We note Class I sources with bright emission at 24 μm and very red μm color ( , where is the magnitude in any of the four IRAC bands) as the subclass “deeply embedded protostars” in the table. Table 3 gives a summary of the classification results.

In Figure 7 we show the spatial distribution of young stars (Class I and deeply embedded protostars, and Class II and transition disk candidates) in W5 overlaid on the IRAC 4.5 μm image. Class I and Class II sources are marked in red and green, respectively, and transition disk candidates and embedded protostars with yellow and blue dots, respectively. O stars are labeled and marked with white asterisks.

Figure 7. Source distribution overlaid on the MIPS 24 μm gray-scale image. Red: Class I; green: Class II; yellow: transition disk candidates. O stars are marked with asterisks.

Figure 7 clearly shows that neither Class I nor Class II type objects in W5 are distributed uniformly, but rather (for the most part) in clustered, or filamentary formations. Within the larger cavity of W5W, clusters of Class II YSOs can be seen centered on the O stars HD 17505, HD 17520, and BD +60 586 of the open cluster OCl 364 (Alter et al. 1970). Similarly, in the smaller W5E bubble a cluster of young stars is visible centered on the O star HD 18326. Smaller clusters of Class I and Class II objects are also visible in the large region of bright diffuse emission to the northwest of HD 17505. Based on the higher proportion of Class I relative to Class II stars (quantified in § 3.4), these are probably younger clusters just emerging from their parental molecular cloud.

Extending out from the central clusters of Class II objects are more filamentary groupings of stars. In W5 West, a chain of young stars appears to extend from the cometary bright rimmed clouds at the southern rim of the bubble (object 11 in the survey of Sugitani et al. 1991) up to the clusters around HD 17505 and HD 17520, and through the cluster around BD +60 586 to the bright rimmed clouds on the eastern rim of the bubble. In W5 East a chain of young stars extends east-west from each of the two peaks of AFGL 4029 (Price & Murdock 1983) through the cluster around HD 18326 to the bright PAH ridge that marks the western edge of the bubble.

Numerous protostars (Class I objects) are seen in projection on the cloud rim at the PAH emission/H ii region cavity boundaries. For example, the bright-rimmed clouds that make up the eastern border of W5 East, the bright rimmed clouds at the northeastern rim of W5 West, and the cometary clouds at the southern rim of W5 West all contain several Class I objects. Several isolated cloud remnants and small “elephant trunk” formations within the H ii region cavity also contain Class I or Class II objects—these are listed in Table 4. We give a type to each object to distinguish the different morphologies. These objects are interesting candidates for triggered star formation on small scales; see the discussion in § 4.3. Figure 8 shows several examples of these objects.

Figure 8. Examples of star-forming globules in W5 shown at 8 μm. PAH emission is seen at or near the globule heads. Objects are labeled as in Table 4.

The influence of star formation on its environment is clear to see in both the IRAC (Fig. 2) and MIPS images (Figs. 3 and 4). Bright diffuse emission in the four IRAC filters (Draine & Li 2007) is only present around the edge of the two main H ii region bubbles in roughly ringlike structures, and brightly lit up over the molecular clouds W5NE and W5NW. Presumably, the PAHs responsible for much of the emission in IRAC bands 1, 3, and 4 are destroyed in the harsh ionizing environment of the H ii region. At 24 μm the morphology is similar in the bubbles’ rims, but very different diffuse emission is seen in the interior surrounding the 5 O stars (as labeled in Fig. 3). Seen in numerous massive star-forming regions by Churchwell et al. (2006), Smith & Brooks (2007), and recently Harvey et al. (2008), this additional component of diffuse emission is consistent with the picture of Crété et al. (1999) whereby emission within H ii regions at mid-IR wavelengths is produced by heated larger dust grains ( Å) that survive being destroyed by ionizing radiation. As Draine & Li (2007) have shown, emission in IRAC bands 1 and 4 is much more sensitive to the PAH mass fraction than is emission in the MIPS 24 μm band. Further, 24 μm band emission increases rapidly when the interstellar radiation field (ISRF) rises to times the local value. Thus, a region with low PAH mass fraction but high ISRF can have enhanced 24 μm emission, but will have low IRAC 3.6 and 8.0 μm PAH emission. The morphology of the 24 μm emission suggests that these larger grains in the immediate vicinity of the O stars are blown away by the strong stellar winds.

Class I and II objects are present in varying amounts across W5. Although with our present data set we cannot tell if any one Class I object is younger than a given Class II object, these classes do represent different evolutionary stages in the formation of stars. In W5 we find that in regions where our emission map exceeds K km s^–1, there are on average 0.23 Class I objects for every Class II. In regions with K km s^–1, the ratio is 0.04. This suggests that the stellar clusters associated with significant molecular material in W5 are systematically younger than those in cleared out regions, for example the H ii region cavities.

The varying levels of completeness in each IRAC map as a function of association with the bright, diffuse emission affect our result. As described in § 2, the 90% limiting magnitude decreases by ∼1.5 mag in IRAC channels 1 and 2, by ∼2.5 mag in channel 3 and by ∼3.2 mag in channel 4 in going from the evacuated H ii region to the brightest regions of diffuse emission. Recently, Chavarría et al. (2008) studied the effects of reduced sensitivity in the 5.8 and 8 μm Spitzer bands, combined with the obscuring effect of bright, diffuse PAH emission on the detection of Class I and Class II sources in IRAC surveys. They found that Class II sources are on average fainter than Class I’s and are likely preferentially missed in bright PAH emission regions as a result.

We want to describe in an objective way how the stars in W5 are grouped, as well as address the following question: is there a population of stars that do not belong to identifiable groups that may have formed in isolation, in other words, a “distributed” population? Lacking a data set that describes the masses and dynamics of the stars and gas which would allow us to assign stars to clusters based on gravitational association, we base cluster membership only on spatial arrangement on the sky and so we do not require that these clusters be gravitationally bound. We restrict ourselves to infrared excess sources only to simplify the issue of whether stars are associated with W5. We include the deeply embedded sources and transition disk candidates along with the Class I and Class II objects and exclude stars exhibiting only photospheric emission (Class III sources) and AGN and PAH galaxies, for a total of 2064 objects. The source distribution is shown in Figure 7 and reproduced in Figure 9 without the background image.

Figure 9. Distribution of stellar sources as in Fig. 7. Green points: Class II; red: Class I; yellow: transition disk candidates. Black asterisks mark the location of the known O stars.

Determining a criterion for cluster membership from spatial distributions alone is somewhat arbitrary. In this paper we use the so-called minimal spanning tree (MST) method to identify and characterize clustering in W5 (see recent work by Cartwright & Whitworth 2004; Bastian et al. 2007). Gower & Ross (1969) describe how such a tree is constructed from a list of source positions. To construct the tree, we first generate the network of lines that joins together the positions of all objects in our input list, such that the total length of all lines joining points is minimized and there are no closed loops. Each object is assigned one “branch length,” that is, the projected distance to its nearest neighbor. We plot the distribution of branch lengths in Figure 10.

Figure 10. MST length distribution, showing each object’s branch length ( -axis) vs. its number in the sorted list. Straight line fits (dashed lines) to the short- and long-length parts of the distribution intersect at 0.86 pc.

Distinguishing clustered from distributed objects using the MST requires two things: a break/cutoff length, , and a minimum group size, . A group is then a collection of objects linked by branches shorter than with at least stars (we choose ). We can estimate a value for from the distribution of branch lengths in Figure 10. Following Gutermuth et al. (2008b), we fit straight lines through the long- and short-length portions of the distribution. Where these lines cross defines the break length for clustering. For the IR excess sources in W5 this method yields pc. Using this break length we find 16 groups containing at least 10 stars and a clustered fraction of 70%.

A single value of chosen in this way may not be representative of the varying spatial density of stars across the region and may underestimate the amount of substructure within groups. Altering changes the number of groups and the relative fraction of clustered and distributed sources. As the cutoff length is increased, an increasing fraction of objects qualify as clustered. The number of groups ( ) we find also increases up to a maximum before falling off as groups start to merge, until all groups belong to a single cluster. These trends are shown in Figure 11. An alternate value for the break length can be derived from the peak in the plot of versus increasing . As argued by Battinelli (1991), this method can be thought of as returning a maximum of information from our source distribution. In this case we find a cutoff of pc, , and a clustered fraction of 44.2%. In Figure 12 we compare the groupings produced by the two methods side by side to demonstrate this finding.

Figure 11. MST results. Solid line and left -scale: Clustered fraction of stars (belonging to all groups ≥10 stars) as a function of branch length cutoff. Dashed line and right -scale: Number of groups ( ) containing 10 or more stars identified by the MST algorithm. Note: The coarse steps on the -axis miss the finer detail in ; the true peak is 27 groups at pc.

Figure 12. Top: MST groups as found using the straight-line fit technique: pc, . Stars are colored according to their group for visual identification only. Bottom: MST groups picked out at the maximum in vs. plot, using pc, . In both plots, black dots represent stars not associated with any cluster. Black asterisks mark the location of the known O stars.

Both values of used here are large when compared to the typical separations of stars in nearby regions such as Taurus (Hartmann et al. 2005), where a cutoff length of 0.54 pc would incorporate many objects not associated with any cluster. However, Taurus is a low-mass star-forming region and so the characteristic scales of clustering are likely to be smaller there. In recent surveys of low-mass star-forming regions (Enoch et al. 2007) and the high-mass star-forming regions in Cygnus (Motte et al. 2007), typical star-forming dense cores are ∼0.1 pc in size. However, Motte et al. (2007) also found that massive dense clumps can be much larger, 0.5-0.8 pc, which may be a more relevant scale for cluster and massive star formation than for low-density, more isolated star formation seen in Taurus.

Table 5 summarizes the clustering results from the two methods. The value for the fraction of stars in clusters from both these methods is lower than that reported by Allen et al. (2007) in Spitzer surveys of Orion and Ophiuchus (74%-78%). Carpenter (2000), in his analysis of the Orion A and B, Mon R2, and Perseus molecular clouds with 2MASS, found a range of clustering fractions, different for every cloud, between 56%-100%. This fraction could be as much as a factor 2 lower, depending on how the age of the distributed population is modeled. In W5—at a distance of 2 kpc—we are sampling much less of the IMF than in these nearby regions, which lie within 1 kpc of the Sun. We also do not know the disk-fraction of YSOs in the W5 clusters. Cieza et al. (2007) have shown that a significant fraction of stars lose detectable circumstellar disks (in the mid-IR) on a timescale ∼1 Myr. These objects would not be included in our sample of IR-excess sources.

The mass completeness limit for our survey is dictated by our photometric completeness and the wide span of YSO mid-IR colors. The primary constraint in completeness for the IRAC four-band sample is the 8 μm detection limit, since this has low sensitivity and very bright background contamination from PAH feature emission. A typical photosphere, with at age 2 Myr and , equates to a limiting mass ≈2 (using the evolutionary models of D’Antona & Mazzitelli [1994] and adopting , a typical value for the H ii region cavity; Hillwig et al. [2006]). On the bright diffuse emission the 8 μm detection limit for a photosphere equates to ∼8 . The additional 2MASS+IRAC bands 1 and 2 sample is limited by the -band completeness. We estimate that the 90% completeness in this band is 14.2. Using the models of Baraffe et al. (1998) this equates to a star of 1 at 2-2.5 Myr, at an of 2 and a distance of 2 kpc. Both the 8 μm and -band mass completeness limits are conservative estimates, since YSOs with infrared excess emission will be detected more readily. For example, the color peaks at ∼1.4 for IR excess YSOs in W5. In this case converts to a mass of ≈0.8 at 2 Myr in the cavity. In summary, we estimate that for a young population of 2 Myr age, we are complete to ∼8 for photospheres and ∼4 for typical disk excess objects in the IRAC four-band sample seen against the bright background. In the cavity we are complete to ∼2 for photospheres and ∼0.8 for disk excess objects. In the additional 2MASS sample we are complete to a mass ∼1 .

Our clustering analysis shows that there is a significant distributed population of young stars (30%-57% of IR excess objects). These may have formed in groups or clusters and dispersed through random motions or been ejected at high velocity. Conversely, they may have formed in their current locations in largely unrelated, isolated events.

On a ∼10 Myr timescale, the distributed population may be a natural consequence of dynamical interactions between young stars in groups throughout the original giant molecular cloud. Small groups ( ) have short relaxation times relative to their crossing times and, once the gas is removed, will not remain visible as clusters for much longer than this (Adams & Myers 2001). However, if we assume that the distributed stars are only as old as the oldest O star in W5 (a reasonable assumption, given that we have identified them through their IR excess emission), they will have had <5 Myr to travel.

Extreme dynamical interactions, namely, close encounters between stars in multiple systems, can produce high-velocity runaway stars. Numerical studies show that a given triple system will eject a member within about 100 crossing times (∼30,000 yr; Reipurth 2000). Sterzik & Durisen (1995) calculate similar timescales for close systems consisting of five stars and predict velocities of 3-4 km s^–1 for the ejected stars. Few have been found; Goodman & Arce (2004) find seven high-velocity (>10 km s^–1) candidates in the literature.

Fűrész et al. (2006, 2008) measured the radial velocities of young stars in NGC 2264 and the Orion Nebula cluster, and found dispersions of 3.5 and 3.1 km s^–1, respectively. In W5 the average projected distance of distributed stars from the nearest cluster is 8.5 pc. Assuming a velocity of 3 km s^–1 perpendicular to the line of sight, this distance could be traversed in ∼3 Myr—well within the current upper age limit for W5 (5 Myr; see Karr & Martin 2003). Thus, it is plausible that at least some of the distributed population originated in nearby clusters.

We calculate the star formation efficiency in W5 using the equation Here represents the mass in young stars only, as identified by their infrared excess, so is a current efficiency, averaged over the last few million years. For the entire W5 region we count 2064 young stars and assume an average mass of 1 . To find the gas mass we use the map of W5 described in § 3.1 and estimate the molecular gas column density from the standard relation: protons cm^–2 (see, for example, Bohlin et al. 1978). Assuming that all gas is associated with W5, we convert to a projected two-dimensional mass density at each pixel in the map: pc^–2 (as in Lombardi & Alves 2001). Each map pixel has a size pc. Over our whole survey field, we find a total gas mass of and derive an efficiency of 3%.

We calculate for the clusters identified by our MST treatment in § 3.4. For each of the 16 clusters, we count the number of stars contained, assume a stellar mass of 1 and add up the gas in the corresponding pixels. Inside the H ii region cavity, where the molecular gas has been destroyed, the efficiencies are high (26%-39%). In the rim of W5, where molecular gas remains and embedded clusters are found, is 10%-17%. These latter values are likely a lower limit, as our census of young stars is incomplete in the bright extended emission coincident with the embedded clusters.

Recent results from the c2d survey (N. J. Evans II et al. 2008, in preparation) show that nearby dark clouds where low-mass star formation is occurring have star formation efficiencies of 2%-4% cloud-wide and 15%-20% in clusters, similar to the values we have obtained for the high-mass star-forming region W5. Our results for in the clusters are also comparable to those of Chavarría et al. (2008) for the S255 region.

We consider two mechanisms of triggered star formation relevant to W5. The first mechanism, called radiatively driven implosion (RDI), is the compression of preexisting density enhancements (small cloud clumps or globules) inside and on the boundary of the ionized bubble by the high pressure of the H ii region (for a review, see Klein et al. 1985). Clumps of material visible as bright PAH emission in IRAC bands 1, 3, and 4 with spatially coincident Class I or Class II objects are summarized in Table 4. These include isolated clumps, elephant trunk formations, and bright rims (as shown in Fig. 8). As discussed in Elmegreen (1998), if globule squeezing does occur, its timescale is expected to be short, since an isolated overdensity can be compressed immediately on being engulfed by the H ii region or stellar wind. Thompson et al. (2004) investigated the cometary features on the southern rim of W5 West for signs of star formation triggered via the RDI mechanism (globules 4-6 in our Table 4). They found that the Hα, CO-molecular, and dust morphologies of the three pillars are “reasonably consistent” with the model of Lefloch & Lazareff (1994) for radiatively driven implosion at the early collapse phase. They used near-IR photometry to detect several candidate young stellar objects—these we confirm through our Spitzer photometry as Class I and Class II objects. They estimate that the timescales for duration of UV illumination, shock crossing times across the features, and protostar/YSO ages are all ∼10⁵ yr. Since Thompson et al. (2004) were only able to establish that approximate pressure equilibrium holds in the pillar heads from their data, we need a more detailed study of the gas dynamics in these pillars to determine if they are currently collapsing due to external pressure from the H ii region.

The second mechanism, collect-and-collapse, is the large-scale expansion of an ionized bubble, powered by an overpressurized H ii region and stellar winds. This expansion can drive shock fronts into the surrounding medium, sweeping up a dense ridge or shell which collapses into stars when a critical density of material accumulates. Whitworth et al. (1994) presented an analytical treatment of this process. A shock front forms and gathers material until it is able to fragment and collapse to form stars, at a time and radius given by where is the sound speed inside the shocked layer in units of 0.2 km s^–1, is the central source ionizing flux in units of 10⁴⁹ photons s^–1, and is the initial gas atomic number density in units of 10³ cm^–3.

In any massive star-forming region, presumably some combination of H ii region expansion and stellar-wind forces operate. Stellar winds certainly play a significant role in shaping the morphology of an H ii region only until an age of ∼10⁵ yr. As shown by McKee et al. (1984) and Weaver et al. (1977), weak stellar winds (wind luminosity ≪ , where is the ambient density) are likely to be confined by the H ii region pressure to a smaller bubble around the ionizing source, since gas that is evaporated from clumps of material in the H ii region mixes with the hot stellar wind bubble and can radiate energy away efficiently. Strong winds can overcome this confinement—they create a bubble with size determined by the photoevaporation and displacement of surrounding inhomogeneities in the ambient gas. This then expands along with the H ii region. In estimating the lifetimes of the W5 bubbles, we thus consider only the effects of the H ii region expansion.

As described in § 1, the gross morphology of W5 in the mid-IR is defined by two large, roughly circular rings of bright PAH emission that mark the smaller W5 East (W5E, radius ∼10 pc) and larger, more irregular W5 West (W5W, overall radius ∼22 pc). At the approximate center of W5E lies HD 18326, an O7 V star surrounded by a dense cluster of young stars. It seems reasonable to assume, whether this is a sphere or a ring of gas, that the cluster lies near the true center and is not significantly offset either to the foreground or background. Adopting stellar parameters from Martins et al. (2005) an O7 V star will have ionizing flux . Since we cannot calculate appropriate values for and from our current data sets, we adopt 1.0 for both. The H ii region fragmentation time we calculate is 1.69 Myr at a radius of 5.37 pc. Both Whitworth et al. (1994) and Dale et al. (2007) note that the density in a real region is likely to be lower—this would make the fragmentation time longer and the radius larger by a factor ∼2.

W5W presents a less clear-cut picture. It contains an isolated O8 V star, HD 237019, as well as three dense clusters centered on BD +60 586 (O7.5 V), and the multiple systems HD 17505 [two O7.5 V((f)) stars, an O6.5 III((f)), and an O8.5 V] and HD 17520 (O9 V+Be). See Hillwig et al. (2006) for a more detailed study of the O stars in W5. The relative configuration of these four systems and their relationship to the surrounding diffuse gas (whether in front or behind) is not known. We consider here two simple scenarios and assume that all four objects and the ring of PAH emission are in the same plane. Scenario 1: HD 237019 represents an initial episode of star formation, and its isolation is due to its cluster having had time to disperse. In the Whitworth model, an O8 V star has a fragmentation time of 1.8 Myr and collapse of swept-up gas occurs at a radius of 5.0 pc. In this picture this event triggered the formation of the three dense clusters in W5W which all lie at roughly 6.8 pc from HD 237019. These clusters, and the continuing expansion of the H ii region, then triggered the current, ongoing star formation seen in association with molecular clouds W5NW and W5NE. Scenario 2: the three dense clusters form together—with their present arclike arrangement due to some initial filamentary distribution of molecular material. They then trigger star formation in the remaining molecular material to the north, northeast, and south. The O7.5 V star BD +60 586 by itself has a fragmentation time of 1.75 Myr and collapse occurs at a radius of 5.16 pc. The combined systems HD 17505 and HD 17520 (with at least five O stars; see Hillwig et al. 2006) have a fragmentation time of 1.46 Myr and collapse occurs at a radius of 6.19 pc. In this picture HD 237019 would have been ejected from one of the multiple O star systems. Located at ≈10 pc from the nearest cluster, assuming a velocity entirely transverse to our line of sight of 10 km s^–1, it would take ∼1 Myr to arrive at its current location, consistent with the likely young age of the cluster as a whole. Moffat et al. (1998) found a runaway frequency percentage of 14% among Galactic O stars and included this star as a marginal runaway candidate. Although the most favored creation mechanism for runaways is supernova ejection (Blaauw 1961), ejection from compact young stellar clusters or through binary interaction is also thought to be responsible for some O star runaways (Poveda et al. 1967; Clarke & Pringle 1992). Moffat et al. (1998) found HD 237019 to be only a marginal runaway candidate, given the large uncertainty in its tangential velocity, pc Myr^–1.

There are several observations consistent with the collect-and-collapse mechanism in W5. The low level of nonthermal radio emission suggests that there have been no supernovae during its lifetime (Vallée et al. 1979). The main-sequence lifetimes of the O stars are estimated to be <5 Myr (Karr & Martin 2003). The timescales for triggering presented above are shorter than this. Wilking et al. (1984) found several young embedded OB stars around W5 in the molecular clouds and argued that their locations near to the cloud edge, together with some evidence of magnetic field alignment parallel to the ionization front are suggestive of triggering via external compression. Finally, Nakano et al. (2008) found evidence of an age difference of ∼3 Myr between the cluster around HD 18326 in the W5 East bubble and the young stars in the rim. This is longer than our estimate above of ∼1 Myr, but given the uncertainties in gas sound speed and initial cloud number density , it is an allowable timescale in the collect-and-collapse model.

Karr & Martin (2003) argue that the distribution of young stars in W5 indicates a shorter timescale, more consistent with RDI than with collect-and-collapse. They investigated the locations of IRAS point sources with colors corresponding to YSOs. Their results did suggest a two-generation model, with older stars at the centers of the H ii regions and ongoing star formation surrounding them; however, they found that the number density of young objects peaks ∼5 pc inside the H ii region (as defined by the 6.2K 1420 MHz contour), which corresponds to a triggering timescale of 0.5-1 Myr. Of the 42 IRAS sources from their list that fall within our Spitzer image, we find only 16 in our source list at 24 μm. The remainder may be spurious (they are also not seen in our 70 μm image) and in fact may be small knots of IR emission that masquerade as point sources in the IRAS survey. All are associated with bright PAH emission, including one coincident with globule 8 in Table 4. The peak in IRAS-classified YSOs found by Karr & Martin (2003) is also not upheld by an analysis of our Spitzer sample, which shows that the Class I objects are found predominantly along the cloud rims or at the cloud centers. This distribution, with the YSOs further from the O stars argues for the longer timescale of collect-and-collapse.

Studies of a young embedded object G138.295+1.555 and the UCH ii region G138.300+1.558 in AFGL4029 (W5E) with near- and mid-IR imaging (Deharveng et al. 1997; Zavagno et al. 1999) suggest that collect-and-collapse is not playing a role in star formation here, since the youngest object (the former) is apparently closer to the ionization front than the older, latter object. However, the true three-dimensional configuration of all the objects involved is not exactly known, this may be due to projection effects. The clustering of Class II objects immediately outside the cloud to the west of G138.295+1.555 and G138.300+1.558 certainly suggests that the sequence of star formation here is west to east, i.e., outside-in, away from the ionizing star HD 18326.

Considering our simple model and the results in the literature, it seems plausible that both the RDI and collect-and-collapse mechanisms are at work in W5: RDI on the smaller scale of cometary globules and collect-and-collapse on the larger scale of the H ii region. Detailed investigations (X. P. Koenig et al. 2008, in preparation) of the spatial distributions and relative ages of the YSOs may further constrain the scenarios presented here.