Citizen Scientists Map Massive Star Formation Throughout the Milky Way

www.milkywayproject.org

Matthew S. Povich California State Polytechnic University, Pomona Milky Way Project Zookeeper and “Science Guru” To be a professional astronomer... What the public thinks we do: What we actually do: Percival Lowell and the Maran “Canals”: An Historical Anecdote on the Perils of “By-Eye” Astronomy

Maran “canals” as drawn by Lowell

It seems a thousand pities that all those magnificent theories of Lowell was particularly human habitation, canal construction, planetary crystallisation, and interested in the canals of the like are based upon lines which our experiments compel us to Mars, as drawn by Italian declare non-existent; but with the planet Mars still left, and the astronomer Giovanni imagination unimpaired, there remains hope that a new theory no Schiaparelli, director of the less attractive may yet be developed, and on a basis more solid than MilanPercival Lowell (1855-1916) Observatory, during the opposition of 1877. “mere seeming.” (wikipedia) —J. E. Evans & E.W. Maunder Monthly Notices of the Royal Astronomical Society 19033 • The original Galaxy Zoo was launched in 2007. • The data: One million galaxies imaged by the robotic Sloan Digital Sky Survey Telescope. • The response: >50 million classifications by 150,000 people in the first year! • The outcome: 50 articles (and counting) in professional astronomy journals, and the Zooniverse was born... Galaxy classificaon—A simple problem, difficult for computers

Normal spiral galaxy Red spiral galaxy Ellipcal galaxy

The human eye–brain combinaon is I know my sll the best paern-recognion system Mommy’s in the known universe! voice and face! 5 2008: Hanny van Arkel discovers a “Voorwerp” Interested citizens who are NOT professional scientists can make discoveries reported in leading journals. www.zooniverse.org Transcribe ancient Greek and Lan texts from Papyri unearthed near Oxyrhynchus, Egypt. Help recover worldwide weather observations made by Royal Navy ships around the time of World War I.

Help decide the ﬁnal desnaon for the New Horizons spacecra. Hunt for planets orbing other stars.

Find and measure star-forming nebulae throughout our Milky Way Galaxy. Count and measure craters and boulder ﬁelds on the Moon.

Classify galaxies in Hubble Space Telescope archive images. Spot explosions on the Sun and track their progress across space to Earth. Match the best simulaons of colliding galaxies to images.

Hunt for exploding stars in the newest astronomical images.

Most stars form in massive clusters.*

*So says the prevailing wisdom, anyway—but how massive is “massive”? Hubble Space Telescope view of the Trapezium — heart of the Orion Nebula Cluster High-mass stars like A low-mass star like the Sun… the Trapezium stars… H II region theory 101: Strömgren spheres

Strömgren (1939): H II region expansion under thermal pressure RS0 = (3Q0/[4πnH2aB])1/3 (Spitzer 1978): is the Strömgren radius. R (t) This is a sharp boundary S between ionized and = [1 + (7/4)(cs2t/RS0)]4/7 neutral gas. (cs2 ~ 10 km/s is the sound speed in the ionized gas). (Q0 is ionizing photon rate, nH is H gas density, aB is Case B recombination coefﬁcient.)

Generally, stars earlier than B3 V emit sufﬁcient Q0 to produce detectible Galactic radio H II regions. Wind-blown bubbles Outer shock—“snowplow” Photoionized shell Contact discontinuity Hot! >107 K; X-rays Wind shock

Castor et al. (1975), Weaver et al. (1977), McKee et al. (1984), Koo & McKee (1992), Capriotti & Kozminski (2001), Freyer Not to scale! et al. (2003), Harper-Clark & Murray (2009) Strömgren spheres Wind-blown bubbles

Complications: • Ambient medium is never uniform— “Do not underestimate the clumpiness of the ISM!” • Multiple stars often contribute to ionization. • Ambient medium generally in motion with respect to ionizing star(s). • Dust! (And magnetic ﬁelds.) • Radiation pressure (see Krumholz & Matzner 2009, Draine 2011). No. 2, 2008 IR DUST BUBBLES 1345

Chandra diffuse soft X-rays Contours: VLA 20 cm (0.5–2 keV) continuum IRAC 5.8 µm Diamonds: OB stars MSX 21.3 µm

Spitzer “false-color” IRAC 4.5 µm • stars Fig. 8.—N49 slice at latitude b 0:23, with 20 cm (solid line, magniﬁed [+shocked/ionized gas] 106 times), 24 m(dotted line), and¼À 8 m(dashed line, magniﬁed 5 times). Note that there is no central peak at 24 m, as there is in N10 and N21. IRAC 8.0 µm • PAHs [+hot

dust] 24 m/20 cm dip appears to be the central wind-evacuated cavity MIPS 24 µm • warm dust expected around early-O stars.

4. ANALYSIS Bubble N49 We proposeText the following picture for the IR bubbles: ionized gas with a hot dust component is surrounded by a PDR containing swept-up interstellar gas, PAHs, and dust. The ionized gas is traced by 20 cm free-free emission, and the hot dust within the H ii region is bright at 24 m via thermal continuum emission. The IR bubbles are enclosed by a shell of 8memissiondominated by PAH emission features in IRAC bands 3.6, 5.8, and 8.0 m. The inner face of the 8 m shell defines the PAH destruction radius from the central ionizing star(s). In the following sections, we determine the PAH destruction radii and PDR shell thick- Triggered massivenesses based on 5.8 m/4.5 m and 8.0 m/4.5 mTownsley et al. (2003) flux density ratios. star formaon? Povich et al. (2007) C. Watson, Povich et al. 4.1. PAH Destruction Povich et al. (2007) argued that ratios of IRAC bands that contain strong PAH emission features (8.0 and 5.8 m bands) to the 4.5 m band (which contains no PAH feature) can be used to de- Evere & Churchwell (2010): Dusttermine must be the PAHconnuously replenished destruction radius and define the extentwithin N49. of PDRs around hot stars. This technique was applied by Povich Fig. 7.—N49 24 m(red ), 8Draine (2011): 20 cm shell in N49 shaped by m(green), 4.5 m(blue), and 20 cm (con- radiaon pressure and winds. tours, bottom panel). The white dashed line in the top panel indicates the loca- et al. (2007) to derive both the PAH destruction region in M17 tion of the cross-cut in Fig. 8. and the extent of its PDR because the 8.0, 5.8, and 3.6 m IRAC bands all contain PAH bands, whereas band ratios involving the 4.5 m PAH-free band should be especially sensitive to regions of 5:7 0:6 kpc (Churchwell et al. 2006). At 1.4 GHz it has an containing PAHs. They supported their interpretation of these ra- integratedÆ flux density of 2.8 Jy and an angular radius out to the tios by showing that the 5.8 m/3.6 m ratio does not delineate background of 1.50 ( 2.5 pc; Helfand et al. 2006). To maintain the PDR boundaries. They also presented IRS spectra that proved 48 1 its ionization, 7:8 ; 10 ionizing photons sÀ are necessary, equiv- the disappearance of PAH features within the M17 H ii region. alent to a single O6 V star (MSH05). The radius to the inner face Povich et al. (2007) were unable to use the 8.0 m images of of the 8 m shell is 1.20 (2.0 pc), and out to the background level M17 because the detector was saturated over large regions. We it is 1.70 (2.3 pc). have applied this technique to N10, N21, and N49. The quanti- N49 has a double-shell structure, the outer traced by 8m emis- tative ratios are different from those toward M17 because M17 sion and the inner traced by 24 m and 20 cm emission (see is a much more luminous region, but the principle is the same. Fig. 7). As in N10 and N21, 8memissionenclosesboththe24m Since N10, N21, and N49 do not saturate the 8.0 m detector, we and 20 cm emission. The transition between the 8 m emission are able to use this band in our analysis as well. ring and the 24 m and 20 cm emission ring can be clearly seen Figs. 9–11 show false-color images of the 5.8 m/4.5 m and in the slice at constant latitude in Figure 8. The 20 cm and 24 m 8.0 m/4.5 m band ratios, with accompanying longitude or emission are coincident, and both have a central cavity. The latitude cuts (averaged over 20 pixels) for all three bubbles. We The “Bubble Search” in the Milky Way Project was inspired by the work of undergraduates (including a freshman) at the University of Wisconsin Churchwell et al. (2006, 2007) Bubble Classification and Measurement

BUBBLINGMethodology GALACTIC DISK 767

Fig. 1.—(a) Illustration of a bubble ( N4) where all four examiners agree on the extent, thickness, and eccentricity. There was disagreement on the orientation of the major axis, but this1 was or because 2 theprimary bubble is almost (student) circular. (b) Illustration classifiers, of a bubble (S175) plus where the 2-3 examiners secondary had substantial disagreements on the size, thickness, and eccentricity. The disagreement was caused mainly by the bright emission to the left of the bubble. In both panels, the annular ellipses fitted by examiner 1 are shown in green.classifiers The white, cyan, working and yellow arrows with show therepresentative extent of the major and minor sub-sample. axes of the inner and outer ellipses as measured independently by the other three examiners. The horizontal scale bars represent 20. stars, whereas the 8.0 memissionisfaintorabsentinsidethe tively. The W77 theory suggests that it may not be easy to distin- bubbles, is bright along the shell that defines the bubbles, and guish between possible static versus expanding bubbles by shell generally extends well beyond the bubble shell boundary. We thickness alone. Some images suggest that the 24 m emission is have examined numerous images constructed using this com- mostly located in a thick shell along the inner boundary of the bination of wavelengths and find that Figure 3 is representative 8 m shell (Fig. 3a), but not all. To determine if those bubbles of the distribution of emission with wavelength in the bubbles so whose 24 m emission peaks at the center are static (not expand- far examined. ing) will require high spatial and velocity resolution line images, To explain the observed difference in distribution of the 8 and preferably of both the ionized and molecular gas associated with 24 m emission, we postulate that the 24 mbandisprobably the bubbles. dominated by hot thermal dust emission plus perhaps a small The fact that 8 memissionappearstobeessentiallyabsentin contribution from small grains that are transiently heated due to the central regions of the bubbles (the front and back sides of the their proximity to one or more hot stars. The 8 mbandisprob- shell that defines the bubble contribute some emission) implies ably dominated by the 7.7 and 8.6 mfeaturesoftenattributedto that PAHs are either easily destroyed in the hard radiation field of PAHs. The 24 memissionnearthehotcentralstar(s)isprob- the central star(s) or blown out of the central regions by stellar ably due to silicate grains since small PAHs do not seem to sur- winds. Also, the images clearly show that the 8 memissionex- vive in these environments (see below). According to Li & Draine tends well beyond the obvious dust shell into the photodissocia- (2001), no more than 10% of interstellar Si can be contained in tion regions (PDRs) of the bubbles (Hollenbach & Tielens very small silicate grains, implying an approximately correspond- 1997 and references therein). This strongly supports the notion ing fraction of their contribution at 24 m. that PAHs are destroyed in the vicinity of hot stars and instead Within some of the bubbles, the 24 m emission seems to trace the PDR regions in the neighborhood of hot stars or peak around a point near the center of the bubble, presumably the clusters. central star or cluster responsible for producing the bubble. This 3.2. Bubble Distribution with Galactic Lon itude morphology is not expected if the 24 m–emitting grains are g located in an optically thin shell along the inner boundary of the Tables 2 and 3 contain a total of 322 partial or complete rings 8 m bubble. This raises two important issues: (1) how the dust (bubbles) found in the GLIMPSE survey. This represents an is able to survive the intense radiation field produced by the cen- average of about 1.5 bubbles per square degree in the survey tral star or cluster, and (2) why the dust has not been evacuated area. This is a lower limit due to the selection biases discussed in by the wind(s) of the central star(s). It is difficult to imagine that 2.1; the true density could be substantially higher. The distribu- the bubbles are static. The strong winds of the central O and B tionx of the number of bubbles in 2N5binswithangulardistance stars, along with overpressure from ionization and high temper- (Ál ) from the Galactic center is shown in Figure 4. The distribu- ature, should cause the dust shells to expand. tion shows no significant dependence on angular distance from The shell thickness of dynamically blown bubbles may have the Galactic center in either the north or the south. One might a wide range of values depending on the type of central star, its naively expect the surface density of bubbles to decrease with age, and the density of the ambient ISM. The Weaver et al. (1977, increasing angular distance from the Galactic center because of hereafter W77) theory predicts that the ratio of shell thickness to the decreasing line of sight through the Galaxy. However, there radius increases with both age and ambient interstellar density. At are two selection effects that work against this: (1) more distant a given age, this ratio increases as the stellar wind luminosity de- bubbles are veiled by foreground diffuse emission, basically lim- creases. For example, at an age of 106 yr and an ambient ISM of iting the number of bubbles to those at the near kinematic dis- 3 density 1 cmÀ , the W77 theory predicts a thickness-to-radius ra- tance or P8 kpc (see Table 4); and (2) an increasingly brighter and tio for an O7 V star of 0.07, but for a B3 V star it is 0.29. At an spatially variable background makes bubble detection increas- age of 107 yr, these ratios increase to 0.22 and 0.33, respec- ingly difficult closer to the Galactic center. In the inner Galaxy, the GLIMPSE Survey Areas

Spitzer/IRAC + MIPS Images of “Bubble” Nebulae

Infrared “color code”: STARS DUST and molecules outlining bubbles WARM DUST inside bubbles Bubbles, bubbles, everywhere…

…some broken.

…some breaking…

31 The MWP DR1 Catalog

• 5,106 bubbles in catalog (including 1,362 “small” bubbles); order of mag increase over previous catalogs. • 85% of Churchwell et al. (2006, 2007) bubbles and 96% of Anderson et al. (2011) bubbles rediscovered by MWP. • 86% of radio H II regions in the inner Galaxy (Paladini et al. 2003) match MWP bubbles. Why did the MWP volunteers outdo the pros? (1) Strength in numbers. (2) Lack of bias 6 Simpson et al. (3) More, better data! asset, for example across different zoom levels, the total view counts are summed, such that a cluster of 5 bubbles drawn onto two assets of 50 views each would have a hit rate of 0.05. The hit rate gives a measure of consensus among users that a bubble is present in the data. The bubble’s dispersion is also calculated as the spread 2 2 in coordinates of the individual classifications (!σb + σl , where σ is the variance in the coordinate value). Only bubbles that were seen 50 times or more, and which have hit rates of 0.1 or more, are included in the final catalogue. This ensures that each final, cleaned bubble is a (a) Image: GLIMPSE only, as in CP06 combination of at least 5 individual users’ drawings and was drawn by volunteers at least 10% of the time when displayed on the website.

3.3 Selection eﬀects It is not known how many bubbles exist in the Galaxy, hence it is impossible to quantify the completeness of the MWP catalogue. There will be bubbles that are either not visi- ble in the data used on the MWP, or that are not seen as bubbles. Distant bubbles may be obscured by foreground (b) Image: GLIMPSE+MIPSGAL, from the MWP extinction. Faint bubbles may be masked by bright Galac- tic background emission or confused with brighter nebular structures. Fragmented or highly distorted bubbles present at high inclination angles may not appear as bubbles to the observer. The MWP’s ‘citizen science’ approach creates its own biases, whilst overcoming others experienced in similar studies. By comparison with CP06, this study has many thou- sands of times more eyes scanning each section of the sky, and each section is broken down to an optimal colour stretch, thus improving the chances of seeing bubble-like structures. The majority of the MWP volunteers have no professional (c) Raw User Drawings (‘Heat Map’) bias or expectation as to what constitutes a ‘good’ bubble. MWP volunteers may experience measurement fatigue when classifying assets with many bubbles. They may also suﬀer bright neighbour bias, and fail to draw quite obvious bubbles that are adjacent to very prominent or beautiful examples.

3.4 Small bubbles and other objects In addition to marking elliptical bubbles on images, users are also encouraged to mark the locations of other interesting objects. Users can mark areas using a simple rectangle and (d) Reduced, ‘cleaned’ bubbles. are asked to label them as either a small bubble, green knot, dark nebula, star cluster, galaxy, fuzzy red object or other. Figure 5. Example of raw user drawings and reduced, cleaned The ellipse-drawing tool of the MWP has a lower size result using a sample MWP image. A GLIMPSE-only colour sam- ! limit of a diameter of 20 pixels (0.45 at the highest image ple is included to illustrate the differences in the appearance of zoom level). The Small Bubble category allows users to mark images inspected by CP06 and the MWP users. Image shown is centred at l=18.8◦, b=-0.125◦,withsize1.5◦ ×0.75◦.Inimage(c) bubbles which are too small to draw in detail but which can all user-drawn bubbles are placed, from all zoom levels, withan still be clearly made out. These small bubbles are reduced in opacity of 2.5%. In image (d) reduced bubbles are placed with an asimilarfashiontothemorecomplexellipses.Toproduce opacity 2× their hit rate, such that bubbles with hit rates ≥ 50% the catalogue of small bubbles listed in Table 3 we use only are drawn as a solid white bubbles. Colour figure available online. the small bubbles drawn by users at the highest zoom level (this is the vast majority of those drawn). Small bubbles marked at lower zoom levels are equivalent to larger bubbles at the higher zoom levels. By rounding their locations to the nearest 20 pixels, the drawings are clustered. Our catalogue of 1,362 small bubbles is given in Table 3. Each of these small bubbles was drawn by at least five users and was drawn by

#c 0000 RAS, MNRAS 000,000–000 Milky Way Project DR1 15

MWP Finds Many More (Apparently) Small, but Thicker BubblesMilky Way Project DR1 15

Figure 14. Distribution of angular diameter of MWP large bub- Figure 17. Distribution of ratio of thickness to outer diameter bles – values from figure 6 of CP06, scaled to 10 ,shownas for MWP large bubbles – values from figure 10 of CP06, scaled × dashedMWP line. TheDR1 smallest radius users can draw is 0.45’. The to 10 ,shownasdashedline. × largest image served to users is 1.5 1 . × ◦ 10 x CP06 bles, like the GLIMPSE bubbles, predominantly trace massive star formation. The distribution of bubbles with Galac- tic latitude reflects a low scale height (Figure 13), similar to molecular clouds and the known Galactic OB population (see CP06). The CP06 bubble catalogue contained only 12% of available Paladini et al. (2003) H ii regions. The MWP catalogue contains 86% of the Paladini sources, indicating that the new catalogue is more complete. A bubble is produced around a massive star when an H ii region, driven by thermal overpressure, stellar winds, radiation pressure, or a combination of these feedback mech- Figure 14. Distribution of angular diameter of MWP large bub- Figure 17. Distribution of ratio of thickness to outer diameter bles – values from figure 6 of CP06, scaled to 10 ,shownas for MWP large bubbles – values from figure 10 of CP06, scaledanisms, expands into the surrounding cold ISM, sweeping × dashed line. The smallest radius users can draw is 0.45’. The to 10 ,shownasdashedline. up gas and dust into a dense shell surrounding a low- × largest image served to users is 1.5 1◦. density, evacuated cavity (Weaver et al. 1977; Garcia-Segura × & Franco 1996; Arthur et al. 2011; Draine 2011). The rela- Simpson et al. (2012) bles, like the GLIMPSE bubbles, predominantly trace mas-tive contributions of different feedback mechanisms likely de- sive star formation. The distribution of bubbles with Galac-pend on the properties of the driving star(s), with the most Figuretic 15. latitudeDistribution reflects of thicknesses a low scale of height MWP large (Figure bubbles 13), – similarmassive, early O-type stars combining powerful stellar winds values from figure 9 of CP06, scaled to 10 ,shownasdashedline to molecular clouds and the known× Galactic OB populationwith high UV luminosities producing ‘wind-blown bubbles,’ (see CP06). The CP06 bubble catalogue contained only 12%while lower-mass, late-O and B dwarfs give rise to ‘classi- of available Paladini et al. (2003) H ii regions. The MWPcal’ H ii regions powered by UV photons alone (CWP07; catalogue contains 86% of the Paladini sources, indicatingWatson et al. 2008). Castor et al. (1975) and Weaver et al. that the new catalogue is more complete. (1977) derived analytic solutions for the expansion of stel- A bubble is produced around a massive star when anlar wind-blown bubbles into a uniform low-density medium. H ii region, driven by thermal overpressure, stellar winds, ra-More recent modelling efforts have included the effects of diation pressure, or a combination of these feedback mech-ionising radiation and the stellar winds (e.g. Capriotti & anisms, expands into the surrounding cold ISM, sweepingKozminski 2001; Draine 2011). up gas and dust into a dense shell surrounding a low- The wind-blown bubbles around massive stars produced density, evacuated cavity (Weaver et al. 1977; Garcia-Seguraby these models display the following general structure: & Franco 1996; Arthur et al. 2011; Draine 2011). The relative contributions of different feedback mechanisms likely de- An inner cavity cleared rapidly by a freely flowing hy- • pend on the properties of the driving star(s), with the mostpersonic stellar wind; Figure 15. Distribution of thicknesses of MWP large bubbles – massive, early O-type stars combining powerful stellar winds A high-temperature region of shocked stellar wind ma- values from figure 9 of CP06, scaled to 10 ,shownasdashedline • 6 × with high UV luminosities producing ‘wind-blown bubbles,’terial (T>10 K); while lower-mass, late-O and B dwarfs give rise to ‘classi- Ashellofshocked,photo-ionisedgas(T 104 K); • ∼ cal’ H ii regions powered by UV photons alone (CWP07; An shell of non-shocked, ionised gas (T 104 K). • ∼ FigureWatson 16. Distribution et al. 2008). of eccentricities Castor et al. of MWP (1975) large and bubbles Weaver et al. An outer shell of neutral material. –valuesfromfigure13ofCP06,scaledto10(1977) derived analytic solutions for,shownasdashed the expansion of stel- • × line. lar wind-blown bubbles into a uniform low-density medium. The bright PAH emission in the PDRs surrounding More recent modelling efforts have included the effects ofH ii regions produces the bright 8 µm bubble rims, while ionising radiation and the stellar winds (e.g. Capriotti &dust mixed with the ionised gas and heated by the hard ra- Kozminski 2001; Draine 2011). diation field produces 24 µm emission interior to the bubbles The wind-blown bubbles around massive stars produced c 0000by RAS, these MNRAS models000 display,000–000 the following general structure: An inner cavity cleared rapidly by a freely flowing hy- • personic stellar wind; A high-temperature region of shocked stellar wind ma- • terial (T>106 K); Ashellofshocked,photo-ionisedgas(T 104 K); • ∼ An shell of non-shocked, ionised gas (T 104 K). • ∼ Figure 16. Distribution of eccentricities of MWP large bubbles An outer shell of neutral material. –valuesfromfigure13ofCP06,scaledto10 ,shownasdashed • × line. The bright PAH emission in the PDRs surrounding H ii regions produces the bright 8 µm bubble rims, while dust mixed with the ionised gas and heated by the hard radiation field produces 24 µm emission interior to the bubbles

c 0000 RAS, MNRAS 000,000–000 The Astrophysical Journal,755:71(15pp),2012August10 Kendrew et al.

Figure 1. Distribution with galactic longitude of the three catalogs: RMS all young sources (green), C06 bubbles (red), and MWP bubbles (blue). Note that the region l 10 was excluded as this is not covered by the RMS catalog. | | ! ◦ (A color versionGlobal of this Association ﬁgure is available of in the IR online Bubbles journal.) with Massive Star Formation (Fortson et al. 2008), in RGB images from the Spitzer Space Telescope GLIMPSE and MIPSGAL surveys (Benjamin et al. 2003;Careyetal.2009). The color composites were created from the 4.5/8.0/24.0 µmimagesoverthecoordinaterange l ! 65◦, b ! 1◦.0. Images were presented online to users, who| | were| asked| to draw the outlines of bubbles using an ellipse-drawing tool. From the inner and outer ellipse sizes, effective radii (Reff)andthicknesses(teff)werecalculatedas simple descriptive metrics for the bubbles, using the equations of C06: 0.5 0.5 (Routrout) +(Rinrin) Reff (1) = 2

0.5 0.5 teff (Routrout) (Rinrin) , (2) = − Kendrew et al. (2012) Figure 2. Galactic latitude distribution of the three catalogs: RMS all young where Rin, Rout are the inner and outer semimajor axes, and sources (green), C06 bubbles (red), and MWP bubbles (blue). Overplotted are rin, rout the inner and outer semiminor axes, respectively (C06; the best-fit Gaussian functions, used for the catalog randomizations. S12). The minimum inner and outer diameters of the drawing (A color version of this figure is available in the online journal.) tool cover 0.#45 and 0.#65, respectively. The minimum Reff corresponding to these values, following Equation (1), is 0.#27. our findings on the prevalence of triggered star formation are The catalog lists inner and outer diameters (or ellipse axes for discussed in Section 5. non-circular bubbles), eccentricities and position angles. Data catalogs for MWP are publicly available online.15 The site also includes a Data Explorer page that visualizes the bubble data. 2. DATA CATALOGS MWP-DR1 consists of two separate catalogs: the large and This section describes the catalogs referenced in the work small bubbles. The small bubbles were not drawn as ellipses presented in this paper. The region of overlap between the RMS, but with simple box shapes. They do not, therefore, have listed C06,andMWPcatalogscovers10◦ ! l ! 65◦, b ! 1◦, and thicknesses in the catalog, nor positional uncertainties. This for each data set sources were selected within| | these| | limits only. study uses the combined large and small bubble sample where Longitude and latitude distributions are plotted for all three it overlaps with the region covered by the RMS sources (see catalogs together in Figures 1 and 2. Section 2.3). The sample contains 4434 bubbles, of which 3260 are “large” and 1174 “small.” Figure 3 shows the distribution 2.1. Milky Way Project Bubbles of Reff of all bubbles, and of the large and small samples individually. To avoid confusion with simple size descriptions, The main focus of the analysis presented in this paper consists we identify these samples as the “MWP-L” and “MWP-S” of the MWP Data Release 1 (DR1) bubbles, presented in S12. bubbles. These bubbles were identified by over 35,000 users of the MWP 14 As noted in S12,severallarge-scalestructuralfeaturesof citizen science Web site, aprojectcreatedbytheZooniverse the Milky Way can be traced in the longitude distribution of

14 http://www.milkywayproject.org 15 http://www.milkywayproject.org/data

3 MWP Bubbles Catalog: Some Science Themes

• “Locally” triggered star formation • Evolution and dynamics of H II regions • Young Cluster Luminosity Functions • Spatially-resolved Galactic star formation rates (SFRs). Possible Mechanisms for “Local” Triggering

• Elmegreen & Lada 1977: Original theory of self- sustaining, sequential massive star formation. • “Radiative implosion” scenario. Clumps of cool, neutral gas enclosed by high-pressure, ionized gas are compressed and form stars (Sandford et al. 1982, Bertoldi 1989). • “Collect-and-collapse” scenario. Gas shells swept-up by expanding I-fronts fragment and form stars (Whitworth et al. 1994). The above are not mutually exclusive. Collect-and-collapse candidates

SuperCOSMOS H-alpha APEX 870 µm Spitzer 8.0 µm Spitzer 8.0 µm

Herschel 250+500 µm RCW 120: “The Perfect Bubble” Zavagno et al. (2007, 2010)

RCW 79 But what happens if Deharveng et al. (2005) bubbles are molecular Zavagno et al. (2006) rings, not spherical shells? (Beaumont & Williams 2010) Mul-generaonal star formaon in giant H II regions: Triggered or regulated by feedback?

Less-evolved young Class I YSOs stellar objects (YSOs) Class II YSOs preferenally located in Transion disks? PDRs, not in central ionizing clusters.

But see also the Carina Complex (Smith, Povich et al. 2010, Povich et al. 2011b)

Koenig et al. (2008) 18 Simpson et al. Bubble Hierarchies

(a) l=332.6◦, b=-0.68◦, zoom=2 (Heat Map) (b) l=332.6◦, b=-0.68◦, zoom=2 (Catalogue)

(e) l=31.8◦, b=-0.06◦, zoom=2 (Heat Map) (f) l=31.8◦, b=-0.06◦, zoom=2 (Catalogue)

Figure 20. Examples of sites showing potential evidence of triggering.Ineachcasethe‘heatmap’andreduceddataisshownoverlaid on the MWP image. Coordinates are image centres, image sizes are indicated by the zoom level (zoom). Zoom levels 1, 2 and 3 refer to images of 1.5 × 1◦,0.75 × 0.375◦ and 0.3 × 0.15◦ respectively. Colour ﬁgure available online.

complement these surveys as a tracer of massive star forma- at http://www.milkywayproject.org/authors. We would like tion on Galactic scales. to thank Bob Benjamin for his helpful suggestions. Also outlined is the creation of a ‘heat map’ of star- The Milky Way Project, and R.J.S. were supported by formation activity in the Galactic plane. This online resource The Leverhulme Trust. The ’Talk’ discussion tool used in provides a crowd-sourced map of bubbles and arcs in the the MWP was developed at the Adler Planetarium with Milky Way, and should enable better statistical analysis of support from the National Science Foundation CDI grant : nearby star-formation sites. DRL-0941610. M.S.P. was supported by a National Science Additonal papers are currently being prepared to out- Foundation Astronomy & Astrophysics Postdoctoral Fellow- line catalogues of ‘green knots’, dark nebulae, star clusters, ship under award AST-0901646. galaxies and ‘fuzzy red objects’ that have also been created C.J.C. is supported by an NSF Astronomy and Astro- by the MWP’s community of citizens scientists. Similarly, we physics Postdoctoral Fellowship under award AST-1003134. anticipate a second, reﬁned bubble catalogue incorporating Support for the work of K.S. was provided by NASA through not only better data-reduction techniques but also 100,000s Einstein Postdoctoral Fellowship grant number PF9-00069 more bubble drawings by volunteers. issued by the Chandra X-ray Observatory Center, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-03060. ACKNOWLEDGMENTS This work is based on observations made with the This publication has been made possible by the partici- Spitzer Space Telescope, which is operated by the Jet pation of more than 35,000 volunteers on the Milky Way Propulsion Laboratory, California Institute of Technology Project. Their contributions are acknowledged individually under a contract with NASA.

"c 0000 RAS, MNRAS 000,000–000 18 Simpson et al.

(a) l=332.6◦, b=-0.68◦, zoom=2 (Heat Map) (b) l=332.6Bubble Hierarchies◦, b=-0.68◦, zoom=2 (Catalogue)

(e) l=31.8◦, b=-0.06◦, zoom=2 (Heat Map) (f) l=31.8◦, b=-0.06◦, zoom=2 (Catalogue)

"c 0000 RAS, MNRAS 000,000–000 Triggered Star Formation: A Statistical Approach

4 An overdensity of massive YSOsC06 around bubblesSpitzer mid-IRversus bubbles Red3 MSX Source (RMS) Catalog our histogram in Fig. 2 is limited to 0.25 bubble radii by the need to obtain sufficient RMS YSOs in each bin. At this resolution the worst-case error in radius (i.e. between mean radius !R" and the semi-major axis a)isslightlylargerthan the width of one bin in Fig. 2. We further subdivided our RMS YSO sample into its constituent YSO and UC HII sub-samples to investigate trends in the separate distributions of YSOs and UC HIIs, for example in evolutionary status versus radius. We found that there is no significant difference between the two subsamples, the histograms of separate YSO and UC HII subsamples are indistinguishable from the combined RMS YSO sample. Again, this may be due to the limited sample statistics that we currently have, or this may indicate that gra- dients in evolutionary status around the bubbles are either not present or, if present, are on smaller angular scales than resolved by our study. In order to confirm this result we also carried out the Figure 1. Three colour GLIMPSE image of a Spitzer bubblesame analysis on the Robitaille et al. (2008) catalogue of in- (N109 from the Churchwell et al. 2006 catalogue), showing the trinsically red sources selected from the Spitzer GLIMPSE strong extended 8 µmPAHand24µmdustemissiontracingthesurvey. The Robitaille catalogue contains a much greater bubble rim. The colour coding in the image is 24 µm(red),8µmnumber of objects than the RMS survey, though at the ex- (green) and 4.5 µm(blue).The24µmimageistakenfromthe Figure 2. Histogram of the number counts of RMS YSOs (com- pense of contamination by an uncertain fraction of AGB Figure 3. Histogram of the number counts of Robitaille et al. MIPSGAL survey (Carey et al. 2009). N109 is one of the largest prising YSO and UC HII classifications from the RMS database) stars (Robitaille et al. 2008). The surface density of objects (2008) Intrinsically Red Objects as a function of angular distance bubbles in the Churchwell et al. (2006) catalogue, with a mean as a function of angular distance from the centre of Spitzer bub- ! from the Robitaille catalogue is shown in Figure 3 and dis- from the centre of Spitzer bubbles.Thompson The distance is et expressed al. (2012) in radius of 14.8 ,andisalsothesiteofnumeroussmallerbubbles. bles. The distance is expressed in terms of normalised bubble The positions of two objects from the RMS YSO sample are in-playsradius. a similar The distribution number counts to are the scaled RMS by YSO the area sample, of the with annulus terms of normalised bubble radius. The number counts are scaled dicated by green circles. ahighersurfacedensitytowardsthebubblesthatsharplycorresponding to each bin and thus represent a surface density of by the area of the annulus corresponding to each bin and thus dropsRMS off to YSOs. a uniform Error bars background are determined level. via Poisson statistics. represent a surface density. Error bars are determined via Poisson The background level is much higher than the RMS statistics. ulate on the likely origin of this star formation and show YSO sample, as would be expected due to the higher surface that it is likely that the bubbles predate the formation of densityrather of Robitaille than arcminutes. et al. (2008) Each intrinsically bin represents red an sourc annuluses the RMS YSOs. Finally in Sect. 4 we present a summary of around the centre of each bubble. The surface area of the compared to the RMS catalogue (Urquhart et al. 2008). 20 ! l ! 186, and so only the bubbles in the south- our conclusions and results. annulus thus naturally increases with increasing radius and The distribution of Robitaille et al. (2008) objects does not ern GLIMPSE survey region are presently covered by the so to obtain the surface density of RMS YSOs we scale the peak at 1 bubble radius, but instead exhibits a relatively MMB survey. The individual masers in the MMB catalogue counts in each bin by the surface area of each correspond- flat distribution out to 1 bubble radius. As the RMS cat- have had their positions interferometrically determined to 2 THE STAR FORMING ENVIRONMENT OF ing annulus. A histogram of these scaled number counts is alogue is constrained to star forming objects at an early sub-arcsecond precision and the maser detections are re- SPITZER BUBBLES shown in Fig. 2. evolutionaryFig. state 2 shows (YSOs a clear and peak UC HII in the regions), number whereas of RMS the YSOs ported in Caswell (2009), Green et al. (2009), Caswell et al. 2.1 The surface density of YSOs Robitailleat a etseparation al. (2008) of catalogue 1 normalised is not, bubble this radius. may indicate At greater (2010), Green et al. (2010), Caswell et al. (2011) and Green the presence of an evolutionary gradient across the bubbles. et al. (2012, in press). We plot the surface density of 6.7 As a first approach to studying the distribution of RMS angular radii the number of RMS YSOs falls sharply, reach- Further classification of the Robitaille et al. (2008) sample GHz MMB masers around the southern Spitzer bubbles in YSOs around Spitzer bubbles we simply measured the num- ing a constant background level of ∼ 3sourcesby2bubble and investigation of their star-forming nature are required Fig. 4. ber counts of RMS YSOs expressed as a function of angu- radii. Within an angular radius of 2 normalised bubble radii to prove this hypothesis. Fig. 4 displays a very similar distribution of masers to lar separation from the bubble centres. Our sample of RMS the number of RMS YSOs is demonstrably higher than at We must also explore the possibility that there may YSOs is comprised of the objects classified as either YSO or angular radii greater than 2 bubble radii. The surface den- that of RMS YSOs and Robitaille et al. (2008) intrinsically UC HII in the RMS database2,seeUrquhartetal.(2008)be ansity intrinsic of YSOs bias projected in the against distributionSpitzer ofbubbles both isthe thus RMS higher red sources, albeit with larger error bars due to the smaller for a description of the RMS database and its classification and Robitaillethan regions et al. external (2008) to catalogues the bubbles, around with a the clearSpitzer peak in sample size. There is a clearly distinguished peak in the system. YSO and UC HII sub-classifications both representbubblesthe due surface to the density common projected mid-infrared against the bands the rimsused of to the maser distribution at an angular offset of 1 bubble radius young, recently formed and predominantly massive stellardetectbubbles. both the bubbles and RMS/Robitaille objects. Al- and the surface density of 6.7 GHz masers drops to a roughly objects that enable us to trace the distribution of recentthough theThe bubblesSpitzer arebubbles principally are relatively identified elliptical, via their with ex- typ- constant background level beyond an offset of 2 bubble radii. star formation around the bubbles. Hereafter we refer totendedical PAH eccentricities emission between at 8 µmandtheRMSYSOsand 0.6 and 0.7. As we normalise by The peak in the surface density of 6.7 GHz masers appears this combined population as the RMS YSO sample. WithinRobitaillethe mean intrinsically bubble radius red sources!R" this are will predominantly have the effect po thatint we to be broader than the corresponding peak in the surface the area covered by the GLIMPSE I survey (Benjamin et al.infraredincorrectly sources, calculate the complex the true mid-infrared normalised environments radius of each o RMf S density of RMS YSOs, however the lower signal-to-noise of 2003) there are 846 objects within the RMS YSO sample andthe bubblesYSO from may the lead bubble to a biascentre, in potentially the identification broadening of point the ob- the MMB surface density makes it difficult to interpret this 322 bubbles from the Churchwell et al. (2006) catalogue. sourcesserved at their peak rims. in surface We investigate density. The this position possibility angles by of ex the- el- difference as a real effect. In order to account for the different angular radii ofaminingliptical the distribution fits to the bubbles of 6.7 areGHz not methanol listed in masers Churchwell drawn et al. All three independently selected YSO catalogues (RMS, the bubbles we divided the angular separation of each RMSfrom the(2006), Methanol but these MultiBeam measurements (MMB) were Survey kindly (Green made etavailable al. Robitaille et al. (2008) red sources and MMB 6.7 GHz object from a particular bubble by the mean radius of the2009)by around Matt the PovichSpitzer (Povich,bubbles. priv. 6.7 comm.) GHz methanol so that we masers could ex- masers) display very similar surface density distributions amine the effect of using the true radius of the bubble instead bubble (!R",column9inthecatalogueofChurchwelletal.are thought to exclusively trace young sites of massive star and we thus conclude that the increase in surface density of the mean radius. We found that there is no significant dif- 2006), i.e. expressing the angular separation in bubble radiiformation (e.g. Menten 1991), and thus allow us to trace the of YSOs towards the bubble rims is a real effect. Given ference between scaling the distance of the RMS YSOs with distribution of massive YSOs around the bubbles indepen- this similar behaviour between catalogues, and the currently the true bubble radius and the mean radius. This is more 2 dently of their mid-infrared emission. more comprehensive knowledge of the properties of the RMS http://www.ast.leeds.ac.uk/RMS than likely due to the fact that the angular resolution of The MMB survey currently occupies a longitude range YSOs (e.g. Urquhart et al. 2011; Mottram et al. 2011a; between l =186andl =20,i.e.excludingtherange Urquhart et al. 2009a,b), we restrict our further analysis The Astrophysical Journal,755:71(15pp),2012August10 Kendrew et al. different data sets using:

ND D ND R NR D + NR R w(θ) 1 2 − 1 2 − 1 2 1 1 (4) = NR1R2 with the same symbols (Bradshaw et al. 2011), where subscripts 1and2indicatethebubbleandYSOsamples,respectively.Total pair counts were normalized such that:

Σθ ND D (θ) Σθ ND R (θ) Σθ NR D (θ) Σθ NR R (θ). (5) 1 2 = 1 2 = 1 2 = 1 2 The random bubble catalog was generated with randomly distributed longitudes, latitudes, and effective radii. The latitudes were drawn from the best-fit Gaussian latitude distribution (shown in Figure 2), and the effective radii follow the best-fit log-normal distribution, to ensure that no artificial over- or un- derdensities are introduced by the randomization as compared Figure 5. Distribution of distances to the RMS YSOs, for 942 sources with with the data. The longitudes were distributed uniformly within published distances. For those sources with unresolved KDAs (58 sources) the the coordinate coverage area. Random YSOs were generated near distance was selected. similarly in longitude–latitude space. The random catalogs contained a factor of 50 more objects than the corresponding data For simplicity, we refer to the sample as “YSOs;” however, the catalogs, ensuring good sampling of the covered area and suffi- range of object types this represents in the sample may well be cient number counts in each θ bin. of relevance to the interpretation of the results. Bootstrap resampling, implemented via random sampling After selecting those sources in the overlap region with the with replacement of the bubble catalog, was used to estimate bubble catalogs (10 ! l ! 65◦, b ! 1◦.0), 1018 sources sampling errors (Ling et al. 1986). 100 bootstrap iterations were remain. Their distribution| with| galactic| | longitude and latitude is carried out for the analysis. Uncertainties are presented at the shown in Figures 1 and 2,respectively.Thetypeclassifications 1σ level throughout. by the RMS team suggest the following distribution: 51% Given that massive stars are known to form almost of sources are H ii regions, 14% diffuse H ii regions, ∼32% exclusively in clustered environments (Lada & Lada 2003), a YSOs and 2% H ii/YSOs and∼ 1% young/old stars. ∼ comparison between the bubble–YSO correlation and the YSO The RMS∼ catalog contains distances∼ for the majority of auto-correlation is important for the interpretation of the cor- sources. These were determined either from the literature relation function output. This allows us to assess whether any or from observations in NH3,CO/CS, methanol or water observed clustering signal is physically meaningful, or simply maser velocity; where the source forms part of a larger com- areflectionoftheunderlyingYSOclustering.Tothisendwe plex, the distance to the complex was adopted (from similar perform an auto-correlation analysis using Equation (3). measurements or the literature). Out of the 1018 sources in our sample, 74 have no distances in the catalog and for a further 4. RESULTS 58 the kinematic distance ambiguity (KDA) is unresolved; for The positional correlation analysis described above was this latter group we choose the near distance. The distribution carried out for a number of instances of bubble and YSO of YSO distances is shown in Figure 5. catalogs, to compare to the findings of T12 and assess the physical significance of the correlations observed. The code 3. CORRELATION ANALYSES used to perform the analysis was written in Python, and is FOR CLUSTERING STUDIES publicly available (see Section 7). Anumberofdiagnosticplotswereproducedforeachanalysis Angular correlation functions are a commonly used tool for run, to assess the code performance and provide information the identification of e.g., galaxy clusters in dense fields, where on sensitivities of the method. Figure 6 shows a comparison they allow for the identification of overdensities in source counts of the distributions of data and random catalogs for bubbles compared with a random distribution (e.g., Papovich 2008; in longitude, latitude, and effective radius. Similar plots were Quadri et al. 2008;Hatchetal.2011). The angular correlation produced to check the YSO random catalog distributions. function w(θ)isdefinedastheexcessprobabilityoffinding The second diagnostic plot shown with each analysis is a box two objects, or two types of objects, separated by a distance θ. CP06 Bubbles–RMS plot of the total pair counts, prior to normalization, in each bin This study employs the commonly used Landy–Szalay estimator over the bootstrap iterations (e.g., Figure 7). The plot shows the (Landy & Szalay 1993)forcalculatingthecorrelationfunction Angular Correlation Functions median pair counts in each bin of θ (red horizontal line), the w(θ): boxes span the lower to upper quartiles, and the whiskers show NDD 2NDR + NRR w(θ) − (3) the range 1.5 the inner quartile range. Outlier points beyond The= AstrophysicalN JournalRR ,755:71(15pp),2012August10 these values are× marked with .TheseplotsinformabouttheKendrew et al. × An overdensity of massive YSOs aroundwhereSpitzerθmid-IRis the bubbles separation5 between objects, and N represents the dispersion and skew of the pair counts across the bootstraps. pair counts between data points (NDD), between data and random 4.1. YSO Auto-correlation points (NDR), and between random and random points (NRR). Equation (3)appliestoonesinglesetofsources,thusdescribing Important for the interpretation of the correlation plots that the auto-correlation or the intrinsic clustering properties of the follow is the auto-correlation of the YSO sample, which de- data. The method can be generalized to be applicable to two scribes the intrinsic clustering properties of these sources. The

Figure 4. Histogram of the number counts of MMB 6.7 GHz Figure 5. The angular cross-correlation of the RMS YSO sample masers as a function of angular distance from the centre of Spitzer and the catalogue of Spitzer bubbles as a function of normalised bubbles. The distance is expressed in terms of normalised bubble bubbleThompson radius. et al. (2012) Figure 9. Correlation function between the matchedKendrew sample et of al. bubbles (2012) from radius. The number counts are scaled by the area of the annulus corresponding to each bin and thus represent a surface density. MWP-DR1 and C06 (275 bubbles) and RMS YSOs, compared with that of the Error bars are determined via Poisson statistics. C06 sample. Figure 10. Comparison of position of 275 MWP-C06 matched bubbles. Outliers (A color version of this figure is available in the online journal.) are marked with squares. to the RMS YSO catalogue. Although the Robitaille et al. (2008) red source catalogue has greater sample statistics and (A color version of this figure is available in the online journal.) is likely to be predominantly comprised of YSOs there is a ND1R2 and NR1D2 the counts of real and random catalogues much greater likelihood of contamination by AGB stars and of RMS source-bubble pairs (and vice-versa), and NR1R2 the other non-YSO types than in the RMS YSO catalogue. Sim- counts between two random catalogues of RMS objects andfunction is shown in Figure 9 (black circles). A strong positive 4.3.1. Analysis Checks ilarly, the lower numbers of objects in the MMB 6.7 GHz Spitzer bubbles. As in Sect. 2.1 we scaled θ to the radius correlation is observed out to 1 Reff, with a peak around 1 Reff maser catalogue favours the continuation of our study using of the individual Spitzer bubble in each pair. To avoid in- Anumberofqualitycheckswereperformedpriortothe the larger and much more studied RMS YSO sample. troducing high levels of noise through the randomly gener-observed with a correlation value of 3at4σ ,confirmingthe ated catalogues we performed 50 realisations of each cat- ∼ full analysis to examine potential sensitivities and biases of the alogue, taking the mean of the results to determine ω(θoverdensity). described in T12.Thelowerabsolutevalueofthe 2.2 The angular cross-correlation of bubbles and The errors on ω(θ)werecalculatedbyabootstrappingap- analysis method to input parameters. YSOs correlation and the somewhat lower significance of the peak can proach (e.g. Ghirlanda et al. 2006; Bradshaw et al. 2011).be ascribed to the different binning used in the plot. Beyond As a first consistency test, the analysis was performed with As a refinement of our simple surface density ap- The Spitzer bubble catalogue was divided into 100 randomly only those MWP bubbles which are also present in the C06 proach we also investigated the distribution of RMS chosen bootstrap catalogues each matching the original cata-2bubbleradii,thecorrelationiseffectivelynon-existentor YSOs around the bubbles using an angular two-point logue size (with replacement). The angular cross-correlation catalog. In S12 the catalogs were cross-matched, and the C06 cross-correlation analysis, a technique more commonly was determined for each bootstrap catalogue and the result-slightly negative, which would indicate a relative sparsity of used to determine the clustering properties of galaxies ing 1σ error in ω(θ)isgivenbythestandarddeviationinsources compared with a random distribution. bubble ID is listed in the MWP data catalogs. This cross- (e.g. Smith, Boyle & Maddox 1995; Ghirlanda et al. 2006; ω(θ)fromthe100randombootstrapsamples. matching between the C06 and MWP catalogs reveals more Wang et al. 2011; Bradshaw et al. 2011). The correlation The resulting angular cross-correlation function is plot- The box-and-whisker plots, Figure 7, show the median, spread function defines the probability of finding a population of ob- ted in Figure 5, which reveals almost exactly the same dis-and outliers in pair counts (prior to normalization) in each bin complex associations than a simple one-to-one matching: in jects at a particular angular separation from a different sec- tribution as seen in the surface density distribution shown in some cases a single C06 bubble contains multiple smaller MWP Figures 2–4. The RMS YSO sample is found to be stronglyover the bootstrap iterations for ND D and ND R ,theinstances ond population. Here, we used the RMS YSO sample as our 1 2 1 2 bubbles, in others multiple C06 bubbles are merged into one first population (D1)andtheChurchwelletal.(2006)cata- correlated with the Spitzer bubble catalogue, particularly atthat contain the bootstrapped bubble catalog. As per Section 3, logue of infrared bubbles for our second population (D2). the radius corresponding to the rim of the bubbles where the large MWP bubble. For this exercise we use only those MWP We calculated the angular cross-correlation using the es- correlation peaks. This peak is significant at the 9σ level.sample 1 represents the bubbles (C06 in this case) and sample timator of Landy & Szalay (1993), modified for the cross- The cross-correlation drops sharply beyond the peak at2theRMSsources.Theplotshowsarelativelylargespread 1 bubbles that are associated with one single C06 bubble. In correlation between population 1 and 2 using the equation bubble radius and beyond a distance of 2 bubble radii the the cases where multiple MWP bubbles are associated with of Bradshaw et al. (2011), i.e. cross correlation decreases to essentially zero. This indicatesin pair counts and a clear skewing of the distribution at small that the probability of finding an RMS YSO near a Spitzer the same C06 bubble, we use only the MWP bubble with the ND1D2 − ND1R2 − NR1D2 + NR1R2 separations (bins 1–7), which is a result of the small sample ω(θ)= (1) bubble is markedly greater at an angular radius of 1 bubble NR1R2 closest positional match. Over the relevant coordinate region, radii, and that beyond an angular distance of 2 bubble radiisizes used in this analysis. The low number counts also explain where ND1D2 represents the normalised number counts the RMS YSO population are essentially uncorrelated with this yielded 275 MWP bubbles. at an angular separation of θ of RMS source-bubble pairs, the presence of a Spitzer bubble. the relatively large error bars in Figure 9. Apotentialsourceofdiscrepancybetweenthisresultand Figures 10 and 11 show the difference in coordinates and that of T12 is the different sample of RMS sources used. As radii of these 275 bubbles. A detailed comparison of bubble described in Section 2,weusethepubliclyavailablecatalog parameters in the C06 samples and MWP is also presented in of “all young sources” from the RMS survey. This sample S12. contains YSOs as well as diffuse H ii regions and all evolutionary The median difference in bubble center positions is 0."06 in stages in between. T12 report using a sample of only YSOs and both longitude and latitude. For 90% of the 275 bubbles the UCHII regions, which contains fewer sources than the set used difference is below 0."30 and 0."28 in longitude and latitude, here. Given the high levels of coincidence between bubbles and respectively. Four bubbles, marked with squares in Figure 10,lie diffuse H ii regions, seemingly confirmed by the comparison of more than 10" from their counterparts in C06.Closerinspection bubble-associated and control YSOs, a higher number of H ii reveals their sizes to be discrepant as well, which is likely to regions in our sample may well increase the overdensity in the be a result of the complexities in the associations described YSO counts at the smallest separations. above. The median size ratio of MWP:C06 parameters is 1.02, or 2%, with the full range covering ratios of 0.034–8.3. 90% of MWP bubbles have sizes within 55% of the C06 value. These numbers plotted in Figure 11 indicate that the MWP catalog 4.3. YSO Clustering around MWP Bubbles size is typically somewhat larger than that in the C06 catalog for a given bubble. The differences are small compared with the The analysis was repeated using the MWP bubble data set bubble sizes, with some outliers. as described in Section 2;theYSOcatalogwasidenticalto The resulting correlation function is shown in Figure 9, that used in the C06–RMS analysis. As before, random catalogs with the C06–RMS correlation overplotted for comparison. The were constructed with 50 times the number of sources in the strong peak at θ 1 Reff is not present in the correlation; input catalog, and 100 bootstrap iterations were performed. however, two peaks= are present at separations of 0.8 and

8 The Astrophysical Journal,755:71(15pp),2012August10 Kendrew et al. Correlation functions stratiﬁed by bubble size.

Kendrew et al. (2012)

Figure 14. Normalized distribution functions of bubble sizes for the C06 (red) Figure 15. Angular correlation function for the MWP bubbles and RMS YSOs, with the MWP bubbles divided into subsamples based on size, as indicated in and MWP (blue) samples. Dashed lines indicate the median values, at 1.!21 and 67% ± 3% of MYSOs and (ultra-)compact H II regions appear to be •the plot legend. 0.!61 for C06 and MWP samples, respectively. (A color version of this figure is available in the online journal.) (Aassociated color version of thiswith figure a is bubble. available in the online journal.) • Approximately 22% ± 2% of massive young stars may have formed as a with 50 times the number of sources in the input catalog, and 100 result of feedback4.3.3. Bubble from and YSO expanding Subsamples H II regions bootstrap iterations were performed. The resulting correlations are shown in Figure 13.Theimprovedstatisticsontheanalysis The large number of bubbles in the MWP sample allows resulting from the larger sample size is reflected in the smaller us to explore specific subsamples of bubbles with potentially size of the error bars. meaningful properties. The different correlations and YSO All three correlation functions display a strong clustering source densities for the MWP-S and MWP-L samples, and the signal of the YSOs on scales of <1Reff of the bubbles in the difference in correlation functions between the MWP and C06 sample. The correlations of the full sample and the large bubbles bubbles, invite a closer examination of the role of bubble size in are almost identical, displaying a decrease in correlation from the bubble–YSO correlation. 0 to 1 Reff.TheMWP-ScorrelationwithYSOsappearstobe First, the normalized size distribution functions for the two higher than for the full and MWP-L samples at the smallest bubble samples are shown in Figure 14.Thisclearlyshowsthat separations, consistent with the YSO source density calculations the C06 bubbles are large compared with the full MWP set. The described in Section 4.1. dashed lines in the plot indicate the median values of the size The correlation function is markedly different from the distributions; at 1.!21, the median of the C06 bubbles sizes is C06-RMS result presented here and by T12.However,given twice that of the MWP bubbles (0.!61). A two-sample K-S test 4 the huge increase in bubble sample size this should not be returned a p-value of 10− ,permittingustoruleoutsimple ∼ unexpected. sampling effects for the observed difference. To investigate the The strong clustering signal in the YSO auto-correlation on dependency of the correlation function on bubble size, the MWP small spatial scales, and typical bubble sizes, make it hard to bubble sample was divided into size bins, containing the 50% interpret the observed correlation. The analysis check shown in largest bubbles (>0.!61; 2235 bubbles), the 25% largest bubbles Figure 12 indicates that >10% of 1000 YSOs were required (>1.!26; 1110 bubbles), and the 10% largest bubbles (>2.!25; 448 to be placed along bubble rims for∼ the signal of the overdensity bubbles). The angular correlation function was calculated for to become significant over the intrinsic clustering trend of each of these subsamples, and the result is shown in Figure 15. the YSOs. It is therefore not sufficient to look simply at the Interestingly, the clustering signal at the smallest separation “associated” and “control” groups in the YSOs, as described in decreases and a positive correlation around 1 Reff emerges as Section 4.1;weneedtodeterminespecificallythenumberof the sample increases in size. The decrease in correlation with YSOs located near the bubble rims. increasing size mirrors the lower YSO surface density toward Of the 678 YSOs in the “associated” sample, 225 are located the MWP-L bubbles (Section 4.1). The correlation for the 0.8–1.6 Reff from the center of an MWP bubble. 87% of these are 10% largest MWP bubbles shows a clear overdensity in the classified as compact H ii regions or MYSOs. The percentage 0.8–1 Reff bin. While the absolute correlation value is relatively of H ii regions (55%) is somewhat higher than in the general low, this point, and that for the 1.2–1.4 Reff bin, carry the RMS YSO sample (51%); however, within the Poisson noise on highest statistical significance in the series at 4.4σ and 3.4σ , these number counts this is not a significant difference from the respectively; these values are very similar to those seen in overall YSO-type distribution. The analysis check performed Figure 9. with the “fake” YSO catalog suggests that this proportion of To rule out that this is simply related to the lower number of YSOs (22% of the full sample) placed near bubble rims should sources used in the analysis, we carried out the same correlation be recoverable as an overdensity by the correlation analysis. It analysis with a random selection of 400 MWP bubbles. This is possible that the positive correlation is simply “drowned” out was repeated three times, and while the scatter of points in by the overdensity at the smallest separations. In the following the individual instances can be large, no statistical overdensities section, we examine the correlation between subsamples of appear. We can therefore conclude that the observed overdensity YSOs and bubbles to see if different correlations are observed. around rims of the largest bubbles is real.

10 MWP: CLOUDS

~25% of IRDCs cataloged turn out to be “holes in the sky” —R. Simpson MWP Citizen Science: Next Up

• MWP Bubbles 1.5: Serve images of individual bubbles in catalogue, focus on improving measurements of size, thickness, etc. Currently available to online volunteers. • MWP 2.0/360: Go to higher latitudes and the outer Galaxy using GLIMPSE 3D, Vela–Carina, GLIMPSE 360 images. COMING SOON! • New datasets: Herschel Hi-Gal (CLOUDS, online now), WISE, IPHAS,...? MWP Advanced Science Analysis: (Non-Exhaustive) To Do List

• DR2: Updated catalogue, improved analysis of user drawings. In progress. • Distance estimates to bubbles. • Photometry of bubbles. • Catalogs of other objects identiﬁed by MWP — galaxies, star clusters, protostellar outﬂows. • Training the computers to recognize these structures... Would the MWP NGC 1566 mid-IR bubbles be d = 11.8 Mpc observable in a nearby, external galaxy? Measure IR Spectral Energy Distributions of Galactic H II regions to measure luminosities.

Plot SFR derived from the X-ray + IR “star counts” methods against equivalent Spitzer/MIPS 24 µm luminosity. Significant, systematic discrepancy compared to Calzetti et al. (2007) extragalactic calibration (dashed line).

Chomiuk & Povich (2011) Machine Learning —Chris Beaumont, CfA (predoc fellow from IfA/Hawaii)

✦ Does a square region with a given posion and scale contain a bubble? ✦ Very similar to how images are scanned to detect faces. ✦ So far, achieve about 90% recall and 0.5% false-posive rate. Get involved! Visit www.zooniverse.org and become a citizen scientist. Contact relevant project team members if you are interested in helping analyze new, unpublished data.