THE STELLAR ARCHEOLOGY OF THE M33 DISK: RECENT STAR-FORMING HISTORY AND CONSTRAINTS ON THE TIMING OF AN INTERACTION WITH M31^*,^†

Stars and gas in M33 provide a fossil record of the events that shaped its past. Because of the obvious challenges presented by crowding, much of the previous work on the resolved stellar content of M33 has examined the outer regions of the disk. There are hints that the orbits of disk stars in M33 may have been disrupted, such that the radial age distribution differs from that in more isolated late-type spirals (Gogarten et al. 2010). In fact, it has been known for some time that stars with properties that are consistent with a disk origin appear in the outer regions of M33, as might be expected if the disk was disrupted. Davidge (2003) found that red giant branch (RGB) stars located between 14 and 17 kpc (7–9 disk scale lengths) from the center of M33 have [M/H] ∼  − 1 ± 0.3 dex. This metallicity is higher than expected for stars that belong to a classical metal-poor halo, but is within the range expected for disk objects. A similar situation occurs in M31, where moderately metal-rich stars are distributed throughout much of the outer regions of the galaxy (e.g., Bellazzini et al. 2003).

Davidge (2003), Block et al. (2007), and Verley et al. (2009) find evidence of a large population of C stars in the outer disk of M33. The discovery of a large number of C stars is suggestive of an elevated star formation rate (SFR) 1–3 Gyr in the past, which is the likely time of an interaction with M31 (Putman et al. 2009). There are indications that these stars may belong to a population that is distinct from the inner disk population. Indeed, if—as argued by Block et al. (2007)—the change that is seen in the 8 μm light profile of M33 is driven by the circumstellar dust contents of M giants and C stars, then this is indicative of a change in stellar content. Block et al. (2007) further suggest that the formation of the C stars may be tied to the events that caused the warp in the M33 H i disk at R_GC = 5 kpc.

Barker & Sarajedini (2008) explore the chemical evolution of the outer regions of M33. The chemical evolution deduced from color–magnitude diagram (CMD) morphology does not follow that expected for a closed box; rather, a time-variable rate of infall is required to explain the enrichment history. Barker & Sarajedini (2008) find that the peak rate of gas accretion occurred 3–7 Gyr in the past, during which time 50%–60% of all gas that has fallen onto the disk was accreted. The rate of infall has since plunged, with only 10% of all infalling gas being accreted during the past 3 Gyr. While Barker & Sarajedini (2008) discuss these results in the context of protracted disk assembly, the drop in the infall rate some 3 Gyr in the past coincides with the timing of an interaction between M31 and M33 as predicted by both Putman et al. (2009) and Bekki (2008). A modest infall rate at the present day would be expected if circumgalactic gas was stripped by an encounter with another galaxy.

Past studies have found evidence for systematic radial variations in the star-forming history (SFH) of the M33 disk. Bastian et al. (2007) find that the number of young stellar groups drops suddenly at radii > 4 kpc, which they attribute to a change in the star-forming ISM, and it is only in the central 4 kpc that Park et al. (2009) find clusters with log(t) ⩽ 7.8. The impact of systematic gradients in the SFH are also seen in integrated light. This is demonstrated by Munoz-Mateos et al. (2007), who find a radial gradient in the specific SFR (sSFR) of M33 based on measurements in the UV and IR.

Cioni et al. (2008) examine the distribution of M giants and C stars in M33 and find that the C/M ratio peaks ∼35 arcmin (∼10 kpc) from the galaxy center. The radial variation in C/M is likely a product of gradients in both mean metallicity and age, in the sense that younger mean ages and higher metallicities occur at smaller radii. Using C/M as a metallicity indicator, Cioni (2009) finds a progressive radial decrease in metallicity with increasing radius in M33 out to 8 kpc, while at larger radii the metallicity profile is flat. Cioni (2009) interprets the break in the metallicity profile as an interface between inner and outer disk/halo components.

The spectral energy distribution of the M33 disk at visible wavelengths is consistent with inside-out formation (Li et al. 2004), although the radial age distribution of stars may be affected by processes other than the mechanics of disk assembly. Williams et al. (2009) use deep Advanced Camera for Surveys (ACS) images to examine stellar content over a range of radii in M33. They find that mean age drops as radius increases when R < 8 kpc, but at R > 8 kpc mean age increases with radius. Such a radial age inversion is consistent with simulations of disk evolution (e.g., Roskar et al. 2008; Sanchez-Blazquez et al. 2009), and can be understood in the context of inside-out disk formation coupled with the dynamical evolution of disk stars, which causes the outermost regions of the disk to be populated by stars that formed at smaller radii, but have migrated outward. The most overt signature of such processes at work in M33 is the difference in scale lengths that characterize the distributions of young stars and old stars, in the sense that old stars are distributed over larger spatial scales than young stars (e.g., Verley et al. 2009; Davidge et al. 2011).

Metallicity gradients are a natural consequence of disk assembly in a hierarchal universe (e.g., Colavitti et al. 2009). However, if M31 passed close to M33 then tidal forces may have mixed gas throughout the disk, thereby flattening pre-existing gradients in the ISM. Rosolowsky & Simon (2008) examine the emission spectra of H ii regions that span a range of radii in M33 and find that Δ[O/H]/ΔR ∼ − 0.027 ± 0.012. This is substantially smaller than the mean [O/H] gradient in the 39 disk galaxies studied by Zaritsky et al. (1994), for which 〈Δ[O/H]/ΔR〉 = −0.058 ± 0.009, where the uncertainty is the standard error in the mean. A caveat is that giant H ii regions, upon which many of the measurements in the Zaritsky et al. (1994) compilation are based, may define steeper gradients than those computed from all H ii regions, regardless of size (Magrini et al. 2010).

Rosolowsky & Simon (2008) find considerable scatter about the mean relation between [O/H] and radius in M33, which they attribute to local inhomogeneities in the ISM. An intriguing result is that H ii regions with low [O/H] (and hence presumably low total metallicity) are found at small radii, suggesting that material that has undergone only modest amounts of chemical enrichment is present in the part of M33 that is expected to be the most chemically mature. The presence of such material could be due to the tidally induced mixing of gas from the outer regions into the central regions of the galaxy. Alternatively, Magrini et al. (2010) argue that the comparatively low mean metallicity computed for H ii regions in the central kpc of M33 may be a result of selection effects in surveys that require the detection of [O iii]4363, which is a diagnostic of electron temperature. Still, it is evident from their Figure 19 that the mean [O/H] in the central kpc remains lower than that at 1.5 kpc even if H ii regions that do not have [O iii]4363 detections are included in the sample.

Very young stars might be expected to track the metallicity trends defined by H ii regions. However, blue supergiants (BSGs) in M33 define a steeper radial metallicity gradient than H ii regions (U et al. 2009), with a slope that is consistent with the mean in the Zaritsky et al. (1994) H ii sample. A possible explanation for this difference in gradients is that U et al. (2009) do not find stellar counterparts to the oxygen-poor H ii regions detected at small radii by Rosolowsky & Simon (2008); this may be a consequence of the modest number of BSGs studied.

The ages of H ii regions and planetary nebulae (PNe) differ by at least a few hundred Myr, and so insights into the recent chemical enrichment history of the M33 disk can be obtained by comparing abundance gradients deduced from these types of nebulae. Magrini et al. (2009) find that Δ[O/H]/ΔR = −0.031 ± 0.013 amongst M33 PNe, and this agrees with the gradient measured from H ii regions by Rosolowsky & Simon (2008). Metallicity gradients are expected to steepen with time if there is infall, and Magrini et al. (2009) attribute the lack of evolution of the [O/H] gradient to a low accretion rate during the past few Gyr. Such a drop in infall rate during intermediate epochs is in broad agreement with the conclusion reached by Barker & Sarajedini (2008) from an investigation of CMDs. Alternatively, a static abundance gradient with time could also signal the accretion of gas with low or modest amounts of enrichment, at a fortuitous rate that is sufficient to balance any increase in metallicity that may arise from the recycling of nuclear processed material into the ISM.

Some properties of M33 may not be typical of late-type spiral galaxies in general. For example, when compared with other late-type galaxies, the star-forming activity in M33 appears to be subdued. Indeed, while the sSFR of M33 is close to the midpoint of spiral galaxies in general, it is near the low end of galaxies of similar morphological type (Munoz-Mateos et al. 2007). The radial sSFR gradient in M33 is also steeper than in the majority of galaxies (Munoz-Mateos et al. 2007). This may indicate a low sSFR at small radii, although the circumnuclear stellar content of M33 is similar to that in other nearby late-type spirals (Davidge 2000; Davidge & Courteau 2002).

The gas within 8 kpc of the center of M33 appears to have a sub-critical density for triggering star formation (Martin & Kennicutt 2001), and this is one possible explanation for the low sSFR when compared with other late-type spirals. Considering the low gas density, the chemical properties of M33 are suggestive of a star-forming efficiency (SFE) that is higher than that among other nearby galaxies, and the SFE increases to progressively larger radii (Magrini et al. 2010). That star-forming activity occurs on a large scale in an environment with a low density may be related to the shear rate (Martin & Kennicutt 2001), which influences the frequency of collisions between molecular clouds, or an increase in ISM pressure caused by material falling into the galaxy, coupled with a kinematically hot interstellar environment (Putman et al. 2009). The gravitational field of stars in the disk may also facilitate the collapse of star-forming material (e.g., Verley et al. 2010).

The youngest stars in late-type galaxies play a key role in defining the classical morphological characteristics of these systems, and if M33 has experienced a recent interaction then residual signatures of such an event may be evident in the spatial distribution of these objects. Massive young stars are also brighter than the vast majority of older stars, and so are less prone to crowding in ground-based observations. In the present paper, u* and g' images obtained with MegaCam on the 3.6 m Canada–France–Hawaii Telescope (CFHT) are used to investigate the properties of young stars in M33.

Two specific aspects of young stars in M33 are examined in this paper. First, we investigate the recent SFH of the M33 disk, with emphasis on how it has changed with time and location during the past few hundred Myr. There are considerable uncertainties associated with SFHs that are based on integrated light indicators, and these uncertainties are avoided by employing star counts as a probe of SFH. While area-to-area variations in SFR may occur within a galaxy due to the passage of spiral density waves or stochastic effects, when averaged over long timescales and large areas the SFRs of isolated star-forming disks are expected to be more-or-less constant with time. Aside from dense galaxy clusters, where there may be a high-density ambient intergalactic medium, galaxy-wide departures from a constant SFR in intermediate-mass galaxies are likely triggered by tidal encounters.

Second, we search for young or intermediate-age stars outside of the main body of the M33 disk. One might expect a mixed bag of such objects at large distances from the galaxy center. The outermost regions of the stellar disk may contain stars that migrated from smaller radii (e.g., Sellwood & Binney 2002), while the passage of spiral density waves might generate isolated areas of star formation (Bush et al. 2010). Tidal effects may pull gas from the inner regions of disks, and studies of interacting galaxies indicate that star formation is not uncommon in tidal debris fields (e.g., Neff et al. 2005). Tidal interactions can also re-distribute existing stars over large areas (e.g., Teyssier et al. 2009). If a tidal plume formed due to an interaction between M31 and M33 then young/intermediate-age stars, that either were present in the disk prior to the interaction or formed during the interaction, might be found in the circumgalactic environment.

An important aspect of this study is the near-complete spatial coverage of the M33 disk. The only areas missed in the stellar disk are those that fall in the gaps between detector banks. While crowding is an issue within 2 kpc of the galaxy center, a large fraction of the brightest blue stars are still resolved there with these data (Section 6). Such complete areal coverage facilitates the suppression of stochastic variations in the SFH that may occur over kpc spatial scales. In this sense, the SFH deduced from our data is more akin to what might be derived from, say, fiber spectroscopic studies of distant galaxies, rather than from pencil-beam surveys of distinct areas within nearby galaxies that typically cover only a few kpc².

Observations in u* are another important aspect of this program. The u* filter is well suited to investigations of intrinsically luminous main-sequence stars, not only because they have effective temperatures that places the peak of their spectral energy distributions shortward of 0.4 μm, but also because of the enhanced contrast with respect to the (predominantly red) unresolved stellar body of M33 when compared with filters having longer central wavelengths. Observations in u* are also beneficial when searching for diffuse ensembles of young and intermediate-age stars in the outermost regions of the disk, where the fractional contamination from unresolved background galaxies, the majority of which have red colors (e.g., Davidge 2008; Mouhcine & Ibata 2009), is substantial. An obvious downside to observing in u* is that there is increased sensitivity to reddening variations when compared with filters that have longer effective wavelengths.

The distance modulus computed by Bonanos et al. (2006), μ₀ = 24.92, is adopted for this study. This distance estimate is based on a geometric calibration (the orbital properties of stars in an eclipsing binary), and is in excellent agreement with recent distances computed from the properties of B supergiants and ACS-based observations of the RGB-tip (U et al. 2009). Galleti et al. (2004) compare various distance moduli of M33, and the results are summarized in their Figure 10. While the Bonanos et al. (2006) value is near the upper end of the distance estimates considered by Galleti et al. (2004), it still falls well within the scatter envelope. In any event, the basic results of this study will not change greatly if a distance modulus that is, say, 0.2 mag smaller were to be adopted.

The line-of-sight extinction is assumed to originate in a uniform absorbing sheet in front of M33, and the adopted total extinction is the result of combining estimates of the foreground and internal components. The foreground component is taken from Schlegel et al. (1998), for which A_B = 0.18, while the internal component is that computed for M33 by Pierce & Tully (1992), for which A_B = 0.16; thus, the total extinction is A_B = 0.34 mag. In Section 4 it is demonstrated that this model correctly predicts the mean extinction toward main-sequence stars in M33. Still, it should be kept in mind that dust in M33 is not uniformly distributed. U et al. (2009) find that line-of-sight dust extinctions among bright A and B supergiants vary by ∼0.16 mag in (B − V), with a mean E(B − V) = 0.08. This mean is consistent with the combined foreground and internal reddening values used here. It is demonstrated in Section 4 that reddening variations, as gauged by the width of the main sequence, are not substantial, with ΔE(B − V) ⩽ ±0.1–0.2 throughout much of the M33 disk.

The paper is structured as follows. Details of the observations and the photometric measurements are discussed in Sections 2 and 3. The CMDs of stars and an assessment of reddening variations throughout the disk are presented in Section 4. The recent SFH throughout the disk is investigated in Section 5 by comparing the luminosity function (LF) of main-sequence stars with models. The spatial distribution of young and intermediate-age stars throughout the disk and its immediate surroundings is examined in Section 6. A summary and discussion of the results follows in Section 7.

Photometric standard stars are observed during each MegaCam run, and the calibration information generated from these observations is inserted into the image headers as part of the ELIXER processing. The photometry used in this paper was calibrated using this information. The u* measurements were transformed into u' magnitudes.

Massey et al. (2006) investigated the UBVRI properties of the brightest stars in M33, and their catalog can be used to check our calibration. The spatial coverage of the Massey et al. (2006) survey restricts these comparisons to the Central Field. The comparisons were restricted to only the brightest stars that fall outside of the main body of the disk to minimize the impact of crowding. The Massey et al. (2006) and MegaCam data were recorded during different epochs, and so variable stars will contribute to the scatter between the two sets of measurements.

u' and g' magnitudes for stars in the Massey et al. (2006) catalog were calculated using the empirical transformation relations found by Smith et al. (2002). The CMD of the stars that were used to check the measurements is shown in the top panel of Figure 2. While the sample contains sources with intermediate and blue u' − g' colors, the majority of sources have red u' − g' colors. This is a consequence of restricting the comparison sample to stars outside of the crowded star-forming disk of M33.

Zoom In Zoom Out Reset image size

The difference between the two sets of measurements, Δu', in the sense CFHT–KPNO, is shown as a function of u' − g' in the lower panel of Figure 2. The mean Δu' is 0.04 ± 0.06 mag, where the quoted uncertainty is the formal error of the mean. The Spearman correlation coefficient between Δu' and u' − g' is −0.34, indicating that there is not a significant correlation between these quantities. The scatter about the calibrating relation between u' − g' and U − B in Figure 13 of Smith et al. (2002) is roughly ±0.1 mag. If the four outliers with large |Δu'|, which are probably variable stars, are not considered then the majority of Δu' values fall within the ±0.1 mag dispersion expected based on the scatter in the empirical relation between u' − g' and U − B. We conclude that the good mean agreement between the CFHT and NOAO measurements indicates that the photometric calibration is reasonably sound.

The uncertainty computed for magnitudes measured by ALLSTAR is based on the quality of the PSF fit, but does not account for systematic errors that might result from, say, crowding; thus, it gives only a lower limit to the actual uncertainties in the photometric measurements. Supplementary information about the photometric uncertainties in ensembles of stars, rather than individual sources, can be obtained from artificial star experiments. These experiments have the merit of also providing information about sample completeness.

Main-sequence stars are the primary targets of this investigation, and so the artificial stars were assigned magnitudes and colors that fall along the observed main-sequence ridgeline. As with actual stars on the sky, artificial stars were only considered to be recovered if they were detected in both filters. The photometry files produced in the artificial star experiments were also culled using the -based criteria described earlier.

The incidence of blending and the random errors in the photometry both increase rapidly with magnitude when the completeness fraction drops below 50%, and so the magnitude at which 50% completeness occurs provides a simple metric for gauging the faint end of the useable data range. The distribution of stars in the M33 disk means that the 50% completeness magnitude changes with location in the Central Field, and the artificial star experiments indicate that 50% completeness occurs near u' = 24 at R_GC = 5 kpc (18 arcmin), and u' = 25.3 at R_GC = 9 kpc (32 arcmin). For comparison, 50% completeness occurs near u' = 25.6 throughout the fields that were observed in 2009B.

The photometry was further characterized by comparing measurements made in overlapping sections of the fields. Such comparisons have the potential of revealing sources of error that are not addressed by the artificial star experiments. The difference in u' between sources common to both the NE and NW fields was found to have a roughly constant dispersion when u' < 24, amounting to σ = ±0.03 mag. This is larger than predicted by the artificial star experiments, which predict a dispersion of ±0.02 mag at u' = 22.

Artificial star experiments require an input PSF, and for ground-based observations the PSF typically is constructed from the data. One consequence is that the uncertainties deduced from artificial star experiments do not account for PSF-matching errors, which will be present at some level in all data sets, as it is not possible to characterize the PSF in a perfect way. The NE and NW fields overlap at the edges of the MegaCam science fields where the PSF can vary over small spatial scales, and the constant dispersion in the difference between the photometric measurements in these fields over a range of magnitudes is consistent with the behavior expected from PSF-matching errors.

PSF-matching errors due to spatial changes in the PSF can be quantified by computing aperture corrections, which are the differences between magnitudes measured using large apertures and those determined from PSF-fitting, at various points across a field. The aperture correction will vary across the field if there are changes in the PSF that have not been tracked. We have computed aperture corrections for bright, isolated sources in the section of the NE field that overlaps with the NW field. The aperture correction varies systematically with location across the field, in the sense of increasing in size toward the edge of the MegaCam field. The amplitude of the variation of the aperture correction is 0.06 mag, which is in excellent agreement with the residual dispersion discussed above. We thus conclude that PSF-matching errors introduce uncertainties of a few hundredths of a magnitude to these data.

The disk of M33 contains a number of young stellar complexes, and the photometry of stars in such systems provides an empirical means of assessing the impact of crowding on the photometry in general. To this end, four associations/star-forming complexes—IC 133, A 31, A 116, and A 127—were selected for study. The last three objects are from the catalog of Humphreys & Sandage (1980). It should be emphasized that these are not star clusters in the traditional sense, but giant star-forming complexes of the type discussed by Efremov (1995). The sizes of these complexes may reflect the spatial scales for star formation defined by turbulence (Elmegreen et al. 2003).

These four particular complexes were selected because they are in isolated environments, thereby making their boundaries relatively easy to identify when compared with similar complexes in denser portions of the disk. The geometric center and radius for each complex were estimated by eye, and the CMDs of objects within the extraction regions are shown in Figure 3. These complexes span a range of source densities. Aperture photometry indicates that IC 113, A 31, and A 116 have comparable surface brightnesses, whereas A 127 is roughly 1 mag arcsec² brighter in u' and g'.

Zoom In Zoom Out Reset image size

Even though these complexes are in isolated portions of M33, there is non-negligible contamination from disk stars. To allow the extent of this contamination to be assessed, the CMDs of sources in an annulus surrounding each complex, that subtends the same total area on the sky as the cluster that it abuts, were also constructed, and the results are shown in Figure 3. The number densities of sources near the faint ends of the background CMDs are higher than in the CMDs of the adjacent clusters, demonstrating the diminished impact of crowding in the (comparatively) low density circumcluster environments.

The main sequence is the dominant feature in the cluster CMDs. Because young stellar regions do not have hard boundaries, some fraction of the stars in the background regions are almost certainly cluster members; still, the brightest main-sequence stars occur in the young complexes, rather than the background fields. Sources to the right of the main sequence are BSGs.

The CMDs of the young stellar complexes are compared with isochrones from Girardi et al. (2004) in Figure 4. The models have Z = 0.008, which is close to the metallicity measured for BSGs in M33 by U et al. (2009). There is reasonable agreement between the predicted and observed main-sequence color of A 31 over a large range of u' magnitude. This suggests that the baseline reddening is appropriate for this complex, and that crowding does not appear to affect the photometric measurements.

Zoom In Zoom Out Reset image size

In the case of A 116, the stars with u' < 22 are well matched by the 10 Myr isochrone, but at fainter magnitudes many sources fall redward of the predicted main sequence. Because A 116 is a star-forming region it is possible that the red objects may be stars that have not yet collapsed onto the main sequence. Still, for these to be pre-main-sequence stars the evolutionary tracks of Cignoni et al. (2009) suggest that there must have been a significant star-forming episode in A 116 within only the past few Myr. For comparison, the main-sequence turn-off (MSTO) of A 116 is well matched by the 10 Myr isochrone.

The agreement between the observations and models is not as good for the other two complexes, with the locus of blue stars falling ∼0.1–0.2 mag redward of the zero age main sequence (ZAMS) for a young population. This could indicate that A_V is ∼0.2–0.3 mag higher for these complexes than the baseline value. The effect of adopting a higher line-of-sight extinction is demonstrated in Figure 4, where the dashed line shows the 10 Myr isochrone with A_V = 0.2 mag additional extinction.

While the agreement with the observed color of the main sequence is improved somewhat with a higher extinction, the agreement at fainter magnitudes remains poor, in the sense that the majority of stars still fall to the right of the predicted ZAMS. A comparatively red main sequence near the faint end of IC 133 would occur if this structure contains stars that span a range of ages. Indeed, the main sequence in the CMD of this complex can be matched if it contains stars that formed during the past ∼100 Myr. This timescale falls within the expected lifetime of large star-forming complexes (e.g., Davidge et al. 2011). However, this explanation does not hold for A 127, where the number of stars plunges near u' = 22.

In summary, even though these are relatively dense stellar aggregates, the photometry of stars with u' < 22 tends to be well matched by model isochrones. The one exception is A 127, which has the highest surface brightness of the four complexes studied. However, even in this complex the isochrones pass through the main sequence at u' ∼ 20. These comparisons indicate that the photometry of main-sequence stars with ages ⩽ 10 Myr should be reliable in relatively dense environments of M33, and in Section 6 this is validated by number counts of sources throughout the disk. Of course, in the lower density portions of the disk the faint limit of the data will be such that stars with ages that are well in excess of 10 Myr can be examined.

The (u', u' − g') CMDs of stars in various radial intervals in the Central Field are shown in Figure 5. The distances given in each panel are measured from the center of M33 in the plane of the disk. A disk inclination of 54 deg was assumed (Pierce & Tully 1992).

Zoom In Zoom Out Reset image size

The influence of individual large star-forming complexes are suppressed when stars from large swaths of the M33 disk are combined. The main sequence forms a prominent plume with u' − g' between 0.0 and 0.5 in the R_GC < 10 kpc CMDs, and the diffuse spray of objects to the immediate right of the main sequence is dominated by BSGs. Source confusion increases with stellar density, and this can clearly be seen in the comparatively shallow faint limit in the CMDs for R_GC < 4 kpc. At the bright end, crowding is mitigated to some extent by the blue wavelength coverage of the u* filter, which avoids the peak in the spectral energy distribution of the majority of disk stars. The brightest blue objects in Figure 5 have u' ∼ 17 when R_GC < 8 kpc, and that this peak brightness does not change with radius when R_GC < 8 kpc suggests that the incidence of blending is not significant amongst the brightest sources.

The properties of the CMDs in Figure 5 might be expected to change with radius based on results presented in previous studies. Munoz-Mateos et al. (2007) and Verley et al. (2009) present evidence that the fractional contribution made by young stars to the overall stellar content increases with radius in M33, although this trend ultimately reverses in the outer galaxy (Williams et al. 2009). Bastian et al. (2007) and Park et al. (2009) find that the number of young stellar groups and clusters drops near R_GC = 4 kpc. In contrast, Thilker et al. (2005a) find that the FUV–NUV color is constant across the disk, with a value that is consistent with a flux-weighted mean age in the range 200–300 Myr, suggesting that the young stellar mixture does not change with radius. In fact, there is no obvious change in the properties of the main-sequence stars and BSGs in the R_GC < 8 kpc CMDs in Figure 5.

Thilker et al. (2005a) find that UV emission from the M33 disk drops at a radius of 30 arcmin (R_GC = 8.5 kpc), signaling a rapid decrease in recent star-forming activity. Signatures of such a drop in star-forming activity should also be evident in the (u', u' − g') CMDs. In fact, a marked change in the brightness of the MSTO between the 6–8 kpc and 8–10 kpc CMDs is evident in Figure 5.

The mean color of main-sequence stars in a given magnitude interval changes with radius. This is demonstrated in Figure 6, where the histogram distributions of u' − g' colors for sources with u' between 20.5 and 21.5 in four annuli are compared. The prominent bump in the color distributions is due to main-sequence stars, and the peak of this feature is displaced to progressively redder values with decreasing distance from the center of the galaxy. A least-squares fit to the peaks of the color distributions indicates that Δ(u' − g')/Δ(R_GC) = −0.021 ± 0.004 kpc⁻¹. The recent SFH of M33 does not change with radius in this part of the galaxy (Section 5), and so the change in main-sequence color is not due to a systematic gradient in the SFH. Rather, the radial trend in mean color could be attributed to systematic changes in the amount of extinction. Adopting the reddening law in Table 6 of Schlegel et al. (1998), gradients amounting to ΔE(B − V)/ΔR_GC = 0.015 ± 0.003 kpc⁻¹ and ΔA_V/ΔR_GC = −0.05 ± 0.01 kpc⁻¹ could explain the trends in peak main-sequence color. Davidge (2007) also found evidence for higher levels of extinction among the youngest stars in the central few kpc of the Sc galaxy NGC 2403.

Zoom In Zoom Out Reset image size

The widths of the color distributions do not change with radius when R_GC < 6 kpc. The artificial star experiments predict a dispersion in u' − g' of ±0.04 mag in the 4–6 kpc annulus, whereas the observed dispersion is $\sigma _{u^{\prime }-g^{\prime }} = \pm 0.135$ mag. The width of the main sequence in these observations is thus not dominated by observational errors. A number of factors, such as evolution away from the ZAMS, differential reddening, binarity, and photometric variability will contribute to a main sequence that is wider than expected from observational uncertainties alone. An upper limit to the amount of differential reddening can be obtained by assuming that it is the dominant source of dispersion in u' − g'. After subtracting in quadrature the contribution made by observational errors, we find that ΔE(B − V) ⩽ ±0.10 and ΔA_V ⩽ ±0.3 throughout the inner disk. For comparison, the scatter in reddening found by U et al. (2009) from AB supergiants is ±0.04 mag in E(B − V).

The CMDs are compared with Z = 0.008 isochrones from Girardi et al. (2004) in Figure 7. The isochrones track the locus of main-sequence stars, as expected if the baseline reddening model applies to the majority of young stars throughout the M33 disk. Comparisons with the isochrones further indicate that the youngest stars in the 4–6 kpc interval have ages ⩽ 10 Myr. However, at larger radii the ages of the youngest stars increase with galactocentric distance, such that in the 12–14 kpc interval the youngest stars have ages ∼ 100 Myr.

Zoom In Zoom Out Reset image size

Star counts provide insights into the recent SFR. We consider the statistics of stars with ages ⩽10 Myr—which have masses < 20 M_☉—as these objects are present throughout the central 8 kpc of M33 (Section 6) and are less prone to crowding-related issues than fainter objects. With a global SFR of 0.7 M_☉ year⁻¹ (Blitz & Rosolowsky 2006), 7 × 10⁶ M_☉ of stars will have formed in the past 10 Myr throughout M33. Adopting the Kroupa (2001) initial mass function (IMF) from 0.08 to 100 M_☉, 8 × 10⁵ M_☉ of this will be in stars with masses in excess of 20 M_☉. This translates into ∼10⁴ stars, and such objects will have u' ⩽ 19.5. For comparison, ∼3 × 10³ objects with u' ⩽ 19.5 are seen throughout M33. Some of these bright objects are undoubtedly binaries or unresolved compact star clusters, and so the actual number of massive main-sequence stars will be less than this.

There is roughly a factor of ∼3 difference between the observed number of very bright stars and the number expected if the SFR = 0.7 M_☉ year⁻¹. Possible explanations for this discrepancy are that (1) the SFR has been lower than 0.7 M_☉ year⁻¹ during the past 10 Myr, (2) many very young stars are missing, either because they are in unresolved blends or are heavily obscured, and hence are not detected, and/or (3) the IMF in M33 may be skewed to the production of fewer massive stars than predicted by Kroupa (2001). We consider the first two of these to be more likely than the third, given that model LFs constructed with a Kroupa (2001) IMF match the slope of the observed LF of bright main-sequence stars throughout M33 (Section 5). In Section 7 we argue that an SFR of 0.7 M_☉ year⁻¹ is probably not sustainable in M33 for a prolonged period of time given the mass of the ISM.

The (u', u' − g') CMDs of sources in the NE, NW, SE, and SW fields are shown in Figure 8. The source densities in these fields are lower than in the Center Field, and so the CMDs in Figure 8 extend to fainter magnitudes than is typical of the Center Field. The majority of sources with u' − g' ⩾ 1 and u' > 24 are background galaxies, although core helium burning stars will also occupy this part of the CMD if an intermediate-age population is present. Blue horizontal-branch stars belonging to an old metal-poor component are too faint to be detected with these data.

Zoom In Zoom Out Reset image size

There is a well-populated main sequence in the CMDs of the NE and NW fields, and many of the brightest main-sequence stars belong to the two star-forming complexes in the northern spiral arm of M33 that were studied by Davidge et al. (2011). Bright main-sequence stars are also seen in the CMD of the SW field, although there are fewer stars than in the NE and NW fields. The CMDs of the portions of the NE and SW fields that have the highest stellar densities are compared with Z = 0.008 isochrones from Girardi et al. (2004) in Figure 9. The areas from which sources were extracted to construct the CMDs in Figure 9 are indicated in Figure 17.

Zoom In Zoom Out Reset image size

Whereas the brightest main-sequence stars near the northern edge of the disk follow the 10 Myr isochrone, the majority of stars in the southern part of the disk have ages ⩾10 Myr. The red envelope of the main sequence in both CMDs tracks the redward extent of main-sequence evolution predicted by the isochrones. This is consistent with the majority of blue sources with u' ⩾ 24 in these areas being intermediate-age stars, rather than background objects.

A rich population of objects with u' − g' ∼ 0.4 and u' > 23.5 are seen in the CMDs of all four fields. The photometric properties of these objects are consistent with them being main-sequence stars with ages ⩾ 100 Myr. While in reality a substantial fraction are either blue galactic nuclei or star-forming regions in background galaxies, it is demonstrated in Section 6 that some of these faint blue objects define structures in the outer regions of the M33 disk, and so are probably stars that belong to M33.

The faint limit of these data is sufficient to allow the SFH during the past few hundred Myr to be probed using the u' LF of main-sequence stars. The structural properties of main-sequence stars are subject to fewer uncertainties than those of more evolved objects, making model LFs of main-sequence stars potentially more reliable probes of SFH than the LFs of stars in advanced stages of evolution. The affect of uncertainties in the advanced stages of stellar evolution on SFHs has been investigated by Melbourne et al. (2010), who compare SFHs of the dwarf galaxy KKH 98 deduced from asymptotic giant branch (AGB) and main-sequence stars. These two stellar types yield SFHs that differ significantly during intermediate epochs, and this is attributed to uncertainties in the physics of AGB evolution. Still, our knowledge of main-sequence evolution is not ironclad, and uncertainties in the mechanics of mass loss, coupled with an incomplete knowledge of convection—the dominant source of energy transport in the central regions of massive stars—are some of the deficiencies that affect models of massive stars as they evolve off the ZAMS, but are still burning hydrogen in their core.

Binarity is not an issue when modeling more evolved stages of evolution at visible wavelengths, as the more massive star in a pair evolves from the main sequence first. The light of the primary then swamps that of its less massive (and hence less evolved) companion(s). Binarity is a concern when modeling the characteristics of main-sequence stars at visible wavelengths as brightness differences between the primary star and any companions are much smaller than if one star is highly evolved. However, as long as the binary fraction and mass ratio characteristics do not change with the mass of the primary, the overall shape of main-sequence model LFs should not be affected by binarity. In recognition of this issue, comparisons with model LFs are restricted to LF shape, rather than the absolute predictions of the number of stars predicted by the models.

Model LFs were constructed from isochrones in the Girardi et al. (2004) compilation using routines in STARFISH (Harris & Zaritsky 2001). IMFs with mass function exponents α = −2.7 (Kroupa et al. 1993) and α = −2.3 (Salpeter 1955) for massive stars were considered. The models assume Z = 0.008. There is a radial metallicity gradient in the inner disk of M33 (e.g., U et al. 2009), and models with lower metallicities than this may be more appropriate for comparisons with outer disk LFs. However, the structural properties of ZAMS stars in the range of stellar masses considered here (>3 M_☉) are not sensitive to metallicity.

The model LFs are used as interpretive tools, and no attempt is made to model particular features in the observations. To this end, models were generated for two SFHs that provide useful diagnostic information. One set of models assumes simple stellar populations (SSPs), in which the component stars are coeval and have a single metallicity. While SSP models are strictly appropriate only for star clusters, these models may approximate the bright end LFs of systems that have undergone a single large star-forming event during recent epochs.

The second set of models assumes an SFR that is constant with time. Such an SFH might be expected to dominate in undisturbed disks if stochastic effects are suppressed by averaging over large areas and long time intervals. The metallicities of stars that form during constant SFR conditions will grow steadily with time, as previous stellar generations process and recycle interstellar material. While our models do not account for such enrichment, it is anticipated that this will not significantly affect the results. Indeed, the age–metallicity relation in the M33 disk during the time interval that we explore—the past few hundred Myr—has been flat (Magrini et al. 2009). Even if an age–metallicity relation that is appropriate for the solar neighborhood were to hold, it would result in only marginal enrichment over the ∼200 Myr time interval that is sampled by the stars examined here.

The LFs of objects in the M33 disk with u' − g' between −0.5 and 0.5, which is the color interval that is dominated by massive main-sequence stars, are shown in Figure 10. The faint limit in the 0–2 and 2–4 kpc LFs is markedly brighter than at larger radii owing to the higher stellar densities in these intervals. Source counts from the outer regions of the NE field, where the density of bright stars that belong to M33 should be negligible, have been subtracted to account statistically for contamination from foreground stars and background galaxies. This correction has only a modest impact on the R_GC ⩽ 10 kpc LFs.

Zoom In Zoom Out Reset image size

The shape of the LFs in Figure 10 changes with radius. Williams et al. (2009) conclude that the stellar content of M33 changes significantly near 8 kpc, and Thilker et al. (2005a) find a drop in the level of UV emission and a change in the FUV–NUV color at this point. It is thus worth noting that the character of the LFs in Figure 10 also changes near R_GC = 8 kpc. The LFs of sources with R_GC ⩽ 8 kpc follow a single power law with a common exponent, whereas at larger radii the LFs at the bright end flatten. The number of sources with u' ⩽ 20 also drops near R_GC ∼ 10 kpc, although this may be due to small number statistics.

Three LFs from Figure 10 are compared with the SSP and constant SFR model LFs in Figures 11 and 12. The models have been scaled to match the number densities of objects with $M_{u^{\prime }}$ between −2 and −4. It is apparent from Figure 11 that the SSP models are much flatter than the observed LFs, indicating that the young stars in the M33 disk formed over a range of epochs, and not in a single large-scale event.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In contrast to the SSP models, the constant SFR models in Figure 12 are an excellent match to the 6–8 kpc and 8–10 kpc LFs in the interval $M_{u^{\prime }} \le -2$. The agreement is noteworthy given that no effort was made to tune the models to match the observations. The agreement with the 10–12 kpc LF is much poorer, in the sense that the slope of the observed LF is not reproduced at the faint end, and the fractional contribution made by older stars is higher than expected from a constant SFR in this radial interval. The absence of stars with $M_{u^{\prime }} < -5$ in this radial interval may be due to small number statistics, as only a few are expected in this brightness range under the assumption of a constant SFR.

The ISM of M33 is not distributed symmetrically throughout the galaxy (e.g., Verley et al. 2010), and so region-to-region variations in the SFH of the M33 disk might be expected. To assess the amplitude of possible large-scale azimuthal SFH variations throughout the M33 disk, the LFs of sources in two of the annuli examined in Figures 11 and 12 were constructed using 90 deg azimuthal gathers, as opposed to the 360 deg gather used to construct the LFs in Figure 10. The four resulting LFs in each radial interval are compared in Figure 13.

Zoom In Zoom Out Reset image size

There are only minor quadrant-to-quadrant variations at the faint end of the LFs in Figure 13. This is perhaps not unexpected given that progressively fainter magnitude intervals contain main-sequence stars that formed over wider ranges of epochs than at the bright end, with the result that variations in SFH average out. However, the situation is very different near the bright end. While small number statistics are a factor amongst the brightest stars, in the interval between $M_{u^{\prime }}= -3.5$ and −1.5 the quadrant-to-quadrant differences are larger than predicted by random errors alone. A systematic difference between the LFs of the northern and southern halves of the galaxy is evident in this magnitude interval, in the sense that the source density in the south is ∼0.4 dex higher than in the north. This suggests that the SFR during the past ∼100 Myr in the southernmost part of the disk has been ∼2 × that in the northernmost part. This is consistent with the southern part of the disk having a higher H i concentration than the northern part (e.g., Verley et al. 2010).

The distribution of sources in the Central Field, selected from the (u', u' − g') CMD using the photometric criteria defined in Figure 1 of Davidge et al. 2011 for characteristic ages of 10 and 100 Myr, are shown in Figure 14. The on-sky source distribution is shown in the top row, while the de-projected distribution, approximating the appearance of M33 as if viewed face-on, is shown in the bottom row. Depth effects have only a minor impact on the de-projected distribution given the moderate inclination of M33 to the observer.

Zoom In Zoom Out Reset image size

The star counts drop markedly at 25–30 arcmin radius, or 7–8 kpc, and this is the point at which UV emission also drops (Thilker et al. 2005a). It is also evident from Figure 14 that stars in the 10 and 100 Myr samples have very different distributions, in that much more clustering and spiral structure is evident in the 10 Myr sample than in the 100 Myr sample. The lack of objects in the 100 Myr sample throughout the central regions of M33 is due to crowding.

The uniform distribution of sources in the 100 Myr sample is largely a consequence of the random motions imparted to objects through interactions with giant molecular clouds, which in turn causes them to diffuse from their places of birth with time. Other processes, such as the churning of stellar orbits by spiral density waves (Sellwood & Binney 2002), further blur the stellar distribution, although these are expected to act over time spans of a few disk rotations, and so should not affect the locations of the stars that are considered in the present study, the oldest of which have an age of only ∼ one disk rotation.

Bakos et al. (2008) find that the luminosity-weighted ages of Type II disks (i.e., those that—like M33—have a downward-breaking light profile at large radii) at visible wavelengths is a minimum at the break radius. Based on the color profiles of these systems, they argue that the break in the light profile is a consequence of the luminosity-weighted age profile of the disk, rather than a discontinuity in the disk mass distribution. There is a possible physical driver for this behavior, as cosmologically based models of late-type spiral galaxies find that the ratio of gas to stellar mass peaks at the break radius, fostering a higher proportion of young stars at this point than elsewhere in the disk (Martinez-Serrano et al. 2009).

How do the radial distributions of young and old stars compare in M33? Verley et al. (2009) investigated the light profile of M33 at wavelengths ranging from 3.6 μm to 160 μm. While the light at mid- and far-infrared wavelengths is thermal in origin, and hence traces interstellar and circumstellar material, at 3.6 μm the light originates predominantly from the photospheres of cool stars, and so observations at this wavelength can be used as a proxy for overall stellar mass. With the exception of an upturn in the innermost regions of the galaxy, the 3.6 μm light profile of M33 shown in Figure 2 of Verley et al. (2009) follows a single power law between 1 and 9 kpc; thus, there is no evidence for a break in the stellar mass distribution near 8 kpc.

The number density of objects in the 10 Myr sample is compared with the 3.6 μm integrated light profile from Figure 2 of Verley et al. (2009) in Figure 15. The 3.6 μm relation has been adjusted to account for the slightly larger distance to M33 adopted for the present study. The zero point of the 3.6 μm profile has also been adjusted so that the curve passes through the midpoint of the 10 Myr star counts.

Zoom In Zoom Out Reset image size

The 10 Myr star counts define a trend that is flatter than the 3.6 μm integrated light profile at R_GC < 8 kpc, indicating that the contribution made by young stars to total stellar mass grows with increasing radius in the central 8 kpc. In addition, that the 10 Myr star counts define a more-or-less linear relation for R_GC ⩾ 1.5 kpc suggests that incompleteness is not an issue for these very bright stars throughout much of the M33 disk. The number density of young stars drops near R_GC ∼ 8 kpc, whereas the character of the 3.6 μm profile does not change at this point. The H i distribution of M33 has been mapped over a large area, and there is a localized maximum in H i emission near 7 kpc (e.g., Figure 3 of Corbelli 2003), which is close to the radius at which young stars make the largest contribution to the stellar mass density. The contribution that the youngest stars make to the total stellar mass drops near R_GC = 8 kpc; the radial distributions of young and old stars in the M33 disk are thus consistent with that found in other systems by Bakos et al. (2008).

In Section 5 it was shown that the recent SFH of the M33 disk changes near 8 kpc, in the sense that at smaller radii the SFR during the past few hundred Myr has been continuous, while outside the break radius there has been reduced star-forming activity over the past few tens of Myr when compared with the past few hundred Myr. The drop in the number density of very young stars near R_GC = 8 kpc, coupled with the continuous trend defined by 3.6 μm light, suggests that the SFR at large radii was not elevated during intermediate epochs. Rather, it is more likely that the SFR in the outer disk dropped during the past ∼100 Myr.

The decrease in the number of young stars at R_GC > 8 kpc is probably due in part to the asymmetric distribution of gas in the outer regions of M33. To the extent that H i traces molecular material, the H i distribution mapped by Putman et al. (2009) suggests that large portions of the circumdisk environment are devoid of star-forming material. Another factor is that the density of star-forming material throughout the M33 disk is sub-critical (Martin & Kennicutt 2001), and the probability of breeching the threshold for star formation is low in the outer regions of M33. Finally, the basic properties of star-forming material may change with radius in M33. Evidence for this comes from the physical sizes of molecular clouds in M33, which appear to decrease in size toward larger radii (e.g., Gratier 2010). Such a trend will affect the sizes of star-forming regions at large radii, which in turn may affect the chances of forming massive stars.

We use the star–star separation function (S3F) to investigate stellar grouping throughout the M33 disk. The S3F, which was defined by Davidge et al. (2011), is the histogram distribution of separations between all possible stellar pairings. Davidge et al. (2011) used the S3F to investigate the distribution of main-sequence stars in the peripheral regions of M33, and significant projected spatial coherence was detected among the youngest stars over separations ⩽0.7 kpc. The degree of coherence in the sample of objects studied by Davidge et al. (2011) was found to decrease with increasing stellar age, and it was concluded that star-forming complexes in this environment dissipate over time spans of a few tens of Myr.

The S3Fs of M33 disk stars in two radial intervals are compared in Figure 16. The current investigation is restricted to stars in the 10 and 40 Myr samples because of the substantial level of incompleteness in the 100 Myr sample at small radii. In addition, as we are interested in stellar grouping on kpc and smaller spatial scales, the comparisons are limited to separations ⩽2 kpc, or roughly one disk scale length. The signal in each interval has been normalized to the number of objects with separations between 0.5 and 1.5 kpc (110–330 arcsec).

Zoom In Zoom Out Reset image size

The shape of the S3Fs in Figure 16 differ from those examined by Davidge et al. (2011), in that prominent peaks at moderate separations are absent. Davidge et al. (2011) studied a comparatively small area on the sky in the outer disk of the galaxy that included two large star-forming complexes, so that there was a bias toward small separations. In contrast, the S3Fs in Figure 16 are constructed from stars that cover large swaths of the disk, so that the influence of single star-forming complexes on the S3F are suppressed.

The S3Fs of the 40 Myr sample are steeper than those in the 10 Myr sample at separations between 0 and 200 arcsec (i.e., between 0 and 1 kpc). There is thus greater spatial coherence among stars in the 10 Myr sample with separations <60 arcsec (<250 pc) than in the 40 Myr sample. The reduced signal at small separations in the 40 Myr sample indicates that stars in the disk of M33 become less tightly grouped over timescales of only a few tens of Myr, due to the diffusion of stars from their places of birth.

Structures of a tidal origin lack a dark matter cocoon, and so their morphology can evolve over timescales that are short when compared with classical galaxies. Indeed, while the disruption timescale for structures in the outer regions of galaxies is comparatively long (e.g., Johnston et al. 1996), the prognosis for the long-term survival of tidal structures is poor, with the vast majority expected to dissipate over time spans of only a few hundred Myr (Bournaud & Duc 2006). Still, some structures may survive over Gyr or longer timescales if the interaction geometry is favorable; for example, long-lived tidal features may form if the orbit of the perturber is more-or-less coplanar with the disk of the host galaxy (Bournaud & Duc 2006).

The stellar content of a tidal feature provides a direct means of probing its history. A critical piece of information that can be gleaned from the stellar content is the age of the structure, which in turn indicates when the interaction occurred. A tidal arm that is pulled from a star-forming disk in an interaction that occurred during cosmologically recent epochs should contain stars that span a wide range of ages, reflecting the extended SFH of the original disk. If the donor was actively forming stars before the interaction then some of the youngest stars in a tidal stream will have formed only shortly before the interaction. However, these may not be the youngest stars in the structure, as stars may form for some period of time after an interaction if local density enhancements persist in the gaseous component of the debris field.

Evidence for low-level on-going star-forming activity is seen in extragalactic debris fields. The nearest example is the region between M81 and M82 (e.g., de Mello et al. 2008), even though the interaction occurred a few hundred Myr in the past (Yun et al. 1994). In addition to star-forming structures, the M81–M82 debris field also contains a number of low density young stellar groupings that appear to have formed during the past few tens of Myr, but are not forming stars at present (e.g., Davidge 2008; Mouhcine & Ibata 2009). These objects suggest that there was likely larger scale star-forming activity in the debris field in the not too distant past.

Whether or not stars form depends on the conditions in the debris field, including the density of star-forming material. There is evidence of tidal material in the circumdisk environment of M33. The ridgeline of the H i distribution mapped by Putman et al. (2009) defines an S-shaped morphology, with arms that extend to the north west and to the south east of the disk. However, whereas the H i density throughout much of the M81–M82 debris field is 10²¹–10²² cm⁻² (Yun et al. 1994), comparable densities are found only in the main disk of M33 (Putman et al. 2009). Indeed, the H i density in the arms that emerge to the north west and south east of the M33 disk is an order of magnitude lower than is seen in much of the M81–M82 debris field.

A stellar structure that may coincide with one arm of the H i emission is discussed by McConnachie et al. (2009, 2010). If the stars in this structure are tidal in origin then they presumably were pulled from the disk during the most recent interaction with M31. In fact, simulations of interactions between M31 and M33 discussed by McConnachie et al. (2009) produce structures that are similar to those detected near M33, while also predicting significant warping of the M33 disk. However, McConnachie et al. (2010) measure a mean metallicity [Fe/H] ∼ − 1.6 for RGB stars in this feature, and such a low metallicity argues ostensibly against a disk origin. Still, if a large population of RGB stars with ages less than a few Gyr are present—as would be expected if these stars originated in a star-forming disk environment—then the metallicity estimated from color information will be skewed to low values if the calibration assumes that the RGB stars are all "old."

The distribution of sources with magnitudes and colors that fall within the 100 Myr boundaries defined in Figure 1 of Davidge et al. (2011) is shown in Figure 17. The source distribution is more-or-less uniform over large fractions of the northern and southern fields. A visual examination of individual sources in the 100 Myr sample indicates that many of the objects in these fields are either background galaxies or individual star-forming regions in the spiral arms of moderately distant spiral galaxies that evaded the culling described in Section 3.

Zoom In Zoom Out Reset image size

The substantial contamination by background galaxies notwithstanding, stellar groupings that are associated with the northern and southern ends of the M33 disk are evident. Of particular note is the distinct spur of objects that extends to the north of the spiral arm, which will be referred to as the Northern Spur. Similar features do not emanate from the southern portion of the M33 disk.

The CMDs of sources in the Northern Spur and the Northern Disk, where the latter region is identified in Figure 17, are compared in Figure 18. These structures have very different stellar contents, in that the Northern Spur lacks the very bright main-sequence stars that populate the Northern Disk. The majority of blue sources in the Northern Spur have u' > 23.5, which corresponds to ages in excess of 100 Myr. The Northern Spur thus does not host obvious on-going star formation. Still, the Northern Spur may not be devoid of star-forming material, as the Putman et al. (2009) H i map shows a nose in the H i emission map that coincides with this structure.

Zoom In Zoom Out Reset image size

The outer regions of M33 thus may contain at least one stellar grouping—the Northern Spur—that coincides with a region of H i emission. Given the evidence for an interaction with M31 as recently as 1 Gyr in the past, the Northern Spur might be the remnant of a structure that formed due to the interaction, and in Section 6.5 it is demonstrated that it is part of a much more diffusely distributed structure. Structures of this nature may dissipate over a few disk rotation timescales, or ∼Gyr.

There are indications that extremely diffuse intermediate-age structures lurk in the peripheral regions of M33. While contamination from background galaxies complicates efforts to identify such objects, diffuse stellar aggregates can still be detected using source counts that are averaged over large angular scales. Such structures are hard to detect in unsmoothed star counts.

The distribution of 100 Myr objects in all five fields is shown in Figure 19. These images show star counts in 500 × 500 MegaCam pixel gathers. This bin size corresponds approximately to 0.5 × 0.5 kpc and so is conducive to the detection of objects with spatial scales of a kpc or higher. A low pass filter was applied to the raw number counts to suppress isolated concentrations of objects that occur on scales of a few hundred MegaCam pixels, as visual examination indicates that many of these tend to be star-forming knots in moderately distant background galaxies.

Zoom In Zoom Out Reset image size

In addition to the main body of the M33 disk, four distinct structures are seen in Figure 19. One of these is a low density extension of the Northern Spur. The connection between this structure and the disk is examined further in Figure 20, which shows the distribution of sources in the NE and NW fields, again with 500 × 500 MegaCam binning. The plume that contains the Northern Spur can be traced over ∼25 arcmin, or ∼7 kpc.

Zoom In Zoom Out Reset image size

Three conspicuous extended structures are labeled DO ("Diffuse Object") 1–3 in Figure 19. The surface density of DO 2 is high enough that it can be seen in the upper right-hand corner of the 100 Myr distributions in Figure 14. The compact nature of DO 3 is suggestive of a background cluster of galaxies. However, such a rich galaxy cluster is not obvious in the MegaCam data, and it is significant that DO 3—like DO 2—is located close to structure in the H i isophotes shown in Figure 1 of Putman et al. (2009), which is reproduced in Figure 19.

DO 1 is located near the middle of the NE+NW fields and is due north of M33. It is evident from Figure 20 that DO 1 may be connected to the main body of the M33 disk and the Northern Spur by a diffuse stellar bridge, as the sources counts to the south and south east of DO 1 are higher than to the north or west. A tongue of objects also extends south of DO 1, pointing directly to the M33 disk. DO 1 is ∼0.5 deg east of the ridgeline of the H i arm mapped by Putman et al. (2009), and is at the eastern edge of the area labeled "S2" in Figure 8 of McConnachie et al. (2010). That stars with ages of a few hundred Myr are seen near the McConnachie et al. (2010) structure raises the possibility that it may contain stars that span a range of ages.

The presence of blue stars with ages ∼100 Myr suggests that star formation occurred in DO 1 (as well as DO 2 and 3) in the not too distant past. DO 1 is spread over ∼400 arcmin² (∼30 kpc²) and thus is comparable in extent to areas of UV emission in the outer regions of M83 (Thilker et al. 2005a). Extended ultraviolet (XUV) disks occur in only a fraction of nearby galaxies (e.g., Zaritsky & Christlein 2007), and the so-called Type I XUV disks, in which the UV emission is clumpy rather than smoothly distributed, are likely the result of spiral structure propagating through the outer regions of disks (Bush et al. 2010).

While M33 does not have an XUV disk at the present day (Thilker et al. 2005b), there are hints that M33 may have experienced vigorous star-forming activity at large radii within the past few hundred Myr. There is a relative excess of stars that formed ⩾100 Myr in the past at distances in excess of 8 kpc from the galaxy center (Section 5), and azimuthal mixing of tidal streams may contribute to producing such a diffuse stellar component (e.g., Oh et al. 2008). This being said, at least some of the older stars in the peripheral parts of the M33 disk may not have formed in situ, but may have migrated there as a result of dynamical interactions. Kinematic heating by sources that are external to the disk but still part of the extended dark matter field of the galaxy might also affect the distribution of stars at large radii (e.g., de Jong et al. 2007).

If star formation in DO 1 occurred in situ then M33 may have had an XUV disk ∼100 Myr in the past. Alternatively, DO 1 may be the remnants of a stellar tidal stream, and this is supported by the close proximity to the structure identified by McConnachie et al. (2010). Deep imaging studies provide a means of distinguishing between these possibilities. Both the XUV and tidal source models predict that DO 1 should contain main-sequence stars that are fainter than those found here. However, the XUV and tidal stream models predict different main-sequence LFs. Whereas XUV activity should produce main-sequence stars with an LF that follows that of an SSP, if DO 1 contains disk stars that span a range of ages then the main-sequence LF should depart significantly from that of an SSP.

A centrally concentrated upswing in star-forming activity is one consequence of a tidal encounter with another galaxy. Tidal torques remove angular momentum from gas in the disk, causing it to migrate inward. Elevated levels of star formation in the inner regions of the galaxy are then triggered as the gas density builds and cools. Pre-existing stars in the disk gain angular momentum via interactions with the infalling gas clouds, and move to larger radii, producing an excess of old and intermediate-age stars in the outer disk (e.g., Younger et al. 2007). If there was a recent interaction with M31 then elevated levels of central star formation may have occurred in M33—is there evidence of such activity in the resolved stellar content?

If the disk of M33 is recovering from an uptick in star-forming activity that occurred within the past few hundred Myr, which is the typical timescale for a starburst, then the SFR should decrease as one moves toward more recent epochs. Such a decrease in the SFR is expected as the gas supply is depleted by star formation and/or feedback. Surveys of M33 star clusters suggest that a large peak in star-forming activity may have occurred ∼0.1 Gyr in the past (Chandar et al. 1999). However, the excellent match between the observed LFs and constant SFR models within 8 kpc of the galaxy center indicates that the global SFR has not changed during the past ∼0.2 Gyr. In fact, given that interactions fuel elevated levels of star formation over roughly a disk rotation time (∼0.3 Gyr), the results presented in this paper suggest that an interaction with M31 could not have occurred during the past ∼0.2 + 0.3 = 0.5 Gyr.

Studies that examine older stellar generations provide additional constraints as to when an interaction between M31 and M33 may have occurred. Williams et al. (2009) find that the SFR near the center of M33 peaked during the past 0.5–2.5 Gyr, at which time the SFR jumped by an order of magnitude above the level that prevailed 2.5–10 Gyr in the past. The time at which this jump in the SFR occurred falls within the constraints set by the present-day positions and motions of M31 and M33 (Putman et al. 2009). Williams et al. (2009) also find that (1) the SFR varied with radius, as expected if gas is displaced by an interaction, and (2) the relative amplitude of the increase in the SFR during this same time interval drops with increasing R_GC, also as expected in the context of inward radial gas flows.

The global SFR during the past few hundred Myr has been lower than the long-term average (Williams et al. 2009), and the sSFR of M33 is lower than in most other late-type galaxies (Munoz-Mateos et al. 2007). These results are consistent with the depletion of cool gas throughout M33, as might occur if there was sustained large-scale star formation during intermediate epochs. The substantial jump in the central SFR during intermediate epochs and the subsequent drop in the SFR since that time suggests that M33 may have experienced a modest starburst; however, any starburst activity was not so dramatic as to restrict present-day star-forming activity to the central few kpc, as is seen in M82, where a large fraction of the outer stellar disk has been denuded of gas.

The drop in the infall rate 3 Gyr in the past inferred from the chemical enrichment properties of the outer disk of M33 by Barker & Sarajedini (2008) may be understood in the context of the stripping of the circumgalactic M33 H i reservoir by M31. The drop in infall occurred at a time that is consistent with the interaction timescales estimated by Putman et al. (2009). The disruption of such a circumgalactic reservoir will affect the overall appearance of M33 in two ways.

First, the removal of gas from the outer halo of a galaxy may affect the size of the star-forming disk. Models of ram pressure stripping indicate that star-forming disks contract as the circumgalactic gas reservoir that fuels star-forming material is removed (e.g., Bekki 2009). Of course, the Local Group environment is such that M33 has not been subjected to significant ram pressure stripping. Still, if the M33 gas reservoir was disrupted tidally then the net impact on the star-forming disk may be similar to that produced by ram pressure stripping: the appearance of a compact star-forming disk for an extended period of time (presumably a significant fraction of the Hubble time) until an external gas reservoir is restored. The sharp outer disk boundary at 8 kpc in M33 may be a consequence of a truncated angular momentum content in the circumgalactic gas reservoir.

Second, the removal of an external reservoir that replenishes disk gas will change the SFR and hence the appearance of the galaxy. If the SFR during the past few hundred Myr has been 0.7 M_☉ year⁻¹ (Blitz & Rosolowsky 2006), then 0.7 × 0.2 × 10⁹ = 1.4 × 10⁸ M_☉ of stars have formed in the past 0.2 Gyr. Adopting a 5% star formation efficiency (Inoue et al. 2000), this level of star formation requires 2.8 × 10⁹ M_☉ of gas. This is substantially lower than the present-day H i mass of M33 (Putman et al. 2009). To be sure, gas is recycled throughout the disk; still, the point to be made is that M33 is not gas-rich and the SFR, which has been constant for the past few hundred Myr, will gradually drop as the disk gas supply is depleted. The drop in SFR during the past 10 Myr that is predicted by the star counts in Section 4.2 may signal the start of this process.

The radial distributions of gas and stars provide clues into the evolution of disks. The radial distributions of gas and stars in isolated disks are expected to differ. This is due in part to the disk assembly process, as the angular momentum distribution of accreted gas will differ from that of stars that are already present. However, even in the absence of gas accretion, the distinct dynamical behaviors of stars and gas results in different radial distributions. Because of their small cross-sections when compared with typical stellar separations, stars form collisionless systems, while the relatively large cross-sections of gas clouds make them collisional systems. Stars may also evolve into extraplanar systems that are distinct from the gas distribution, even in isolated galaxies. Indeed, some fraction of disk stars at small radii may evolve into a thick disk, as a natural consequence of inside-out disk formation (e.g., Schonrich & Binney 2009).

The distributions of gas and stars in disks have been modeled, and the results can be compared with the observed distributions in M33. Here, we consider the cosmologically based galaxy simulations examined by Martinez-Serrano et al. (2009). A break in the model light profile typically occurs at ∼3 disk scale lengths. The gas distribution in these models is roughly flat within the break radius and gradually decreases beyond this point. Star formation continues past the break radius, occurring out to 4.3 scale lengths, at which point it ceases due to the low density of star-forming material. The ratio of gas mass to stellar mass peaks at the break radius, making this the point in the disk with the lowest photometrically-weighted mean age.

The H i density profile throughout much of the M33 disk is relatively flat (Corbelli 2003; Putman et al. 2009), in agreement with the Martinez-Serrano et al. (2009) models. The general trend of increasing specific SFR with radius in the main body of the disk predicted by the models is also consistent with the relative contributions made by young and old stars shown in Figure 15. Still, the muted evidence for star formation outside of the break radius in M33 is not consistent with the models, and the presence of tidal features in the H i distribution outside of the disk (Putman et al. 2009) points to an event that undoubtedly affected the evolution of M33. The presence of large-scale tidal features aside, the distributions of gas and stars within the break radius of M33, where the characteristic timescale for relaxation is shorter than at large radii, are at least qualitatively consistent with predictions made for disk galaxies evolving in the cold dark matter cosmological paradigm. That the relative distributions of gas and stars in the main disk of M33 are consistent with that of an isolated disk suggests that sufficient time has passed (i.e., a few disk rotations) to allow the distributions of stars and gas within the disk to regain at least the semblance of an equilibrium state.

An interaction between M31 and M33 would almost certainly have imprinted signatures in the stellar content of the former. McConnachie et al. (2009) suggest that an interaction with M33 would have disturbed the outer disk of M31, causing a warp and displacing disk stars to large radii. An interaction with M33 may also have triggered an inward flow of gas in M31, depleting gas reserves at large radii. Indeed, the gas content of M31 at large radii is deficient when compared with models of isolated evolution (Yin et al. 2009). The interaction may also have produced elevated SFRs in M31, and a sharp drop in overall star-forming activity would occur as gas is consumed during star formation or is disrupted by feedback. Williams (2002) concludes that a global downturn in star-forming activity occurred in M31 ∼ 1 Gyr in the past. Given that starbursts appear to last no more than a Gyr (e.g., Leitherer 2001), this result—if due to an interaction with M33—suggests that M31 and M33 interacted no more than ∼2 Gyr ago, which is consistent with the age ranges estimated by Putman et al. (2009) and McConnachie et al. (2009). Of course, all of these properties of M31 could also be explained by an interaction with another (possibly now defunct) companion.

If there was a close interaction between M31 and M33 then some of the features traditionally associated with M31 companions could instead be remnants of this interaction. Consider the M31 Giant Southern Stream (GSS; Ibata et al. 2001). The simulations run by Bekki (2008) indicate that M33 may leave a gaseous debris stream near M31, and Fardal et al. (2008) conclude that the GSS may have originated from a spiral galaxy. Indeed, the GSS contains stars with a metallicity as high as [Fe/H] ∼ − 0.5 (Guhathakurta et al. 2006), which is consistent with an origin in the M33 disk. The velocity dispersion of stars in the GSS (e.g., Ibata et al. 2004) is also consistent with an origin in a kinematically cold environment, such as a disk. Finally, the GSS is 85 kpc behind M31 (e.g., Tanaka et al. 2010), and simulations suggest that the GSS was produced by an object on a highly eccentric orbit (Font et al. 2006; Fardal et al. 2006, 2007), such as M33.

While some properties of the GSS can be rationalized in the context of an interaction with M33, efforts to link the GSS with such an event run into timing issues. The GSS has a dynamical age of a few hundred Myr (e.g., Font et al. 2006), which is more recent than the minimum probable age for an interaction estimated by Putman et al. (2009), and is not consistent with the time interval during which there has been a constant SFR throughout the M33 disk. Another issue is that the vast majority of stars in the GSS have an age ⩾ 4 Gyr (Brown et al. 2006), and so fall outside of the nominal age range predicted for an interaction. Still, a modest population of young stars is present in the GSS (e.g., Figure 17 of Brown et al. 2006). Recently formed stars in extraplanar environments may not be well mixed with underlying stellar material, as they formed in gas concentrations, whereas pre-existing stars will tend to dissipate spatially. If this is the case in M31 then future observations may find concentrations of intermediate-age stars in different fields than those sampled by Brown et al. (2006).