Three-dimensional geometry and dust/gas ratios in massive star-forming regions over the entire LMC as revealed by the IRSF/SIRIUS survey

Abstract

We derive the entire dust extinction (A_V) map for the Large Magellanic Cloud (LMC) estimated from the color excess at near-infrared wavelengths. Using the percentile method we recently adopted to evaluate A_V distribution along the line of sight, we derive three-dimensional (3D) A_V maps of the three massive star-forming regions of N44, N79, and N11 based on the IRSF/SIRIUS point source catalog. The 3D A_V maps are compared with the hydrogen column densities N(H) of three different velocity components where one is of the LMC disk velocity and the other two are of velocities lower than the disk velocity. As a result, we obtain a 3D dust geometry suggesting that gas collision is ongoing between the different velocity components. We also find differences in the timing of the gas collision between the massive star-forming regions, which indicates that the gas collision in N44, N79, and N11 occurred later than that in 30 Doradus. In addition, a difference of a factor of two in A_V/N(H) is found between the velocity components for N44, while a significant difference is not found for N79 and N11. From the 3D geometry and A_V/N(H) in each star-forming region, we suggest that the massive star formation in N44 was induced by an external trigger of tidal interaction between the LMC and the Small Magellanic Cloud, while that in N79 and N11 is likely to have been induced by internal triggers such as gas converging from the galactic spiral arm and expansion of a supershell, although the possibility of tidal interaction cannot be ruled out.

1 Introduction

The Large Magellanic Cloud (LMC) is a dwarf galaxy with morphological features such as an off-center stellar bar and a single spiral arm, accompanied by the Magellanic Bridge, which is a H i gas structure connecting the LMC and the Small Magellanic Cloud (SMC). These features are considered to be evidence for the tidal interaction between the LMC and the SMC (e.g., Besla et al. 2012; Diaz & Bekki 2012). The LMC is one of the nearest interacting galaxies at a distance of ≃50 $\mathrm{k}\mathrm{p}\mathrm{c}$ (Pietrzyński et al. 2019) with almost face-on orientation (i ∼ 35°; van der Marel & Cioni 2001), which enables us to investigate the spatial distribution of the interstellar medium (ISM) with high angular resolution (<1 pc) in almost two dimensions. In addition, the metallicity of the LMC (Z ∼ 0.5 Z_⊙; Westerlund 1997) is similar to a value typical of the ISM at redshift z ∼ 0.5 (Pei et al. 1999) corresponding to a relatively active phase of the star-formation history in the Universe. Furthermore, active star formation is ongoing in the LMC, where 30 Doradus (30 Dor) is the most active starburst region in the Local Group (e.g., Kennicutt 1984). Thanks to these characteristics, the LMC is a site suitable for investigating how galactic interaction occurs and how massive star formation is induced in low-metallicity environments.

Recent studies using H i and CO data suggest that the massive star formations in 30 Dor and N159 in the LMC H i ridge region were induced by the galactic interaction between the LMC and the SMC (Fukui et al. 2017, 2019; Tokuda et al. 2019). Fukui et al. (2017) identify two velocity H i components; one is the velocity component of the LMC disk and the other is a low-velocity component relative to the disk velocity. Fukui et al. (2017) and Tsuge et al. (2019) identify an intermediate-velocity component between them for 30 Dor and N159, which is considered to be evidence for gas collision. In addition, they evaluate the dust/gas ratio from the comparison of the H i intensity with the dust optical depth, and find that the dust/gas ratio for the LMC H i ridge region is lower than that for the LMC stellar-bar region. Based on the low dust/gas ratio and numerical simulations (e.g., Bekki & Chiba 2007a, 2007b; Yozin & Bekki 2014), they suggest that low-metallicity gas from the SMC (Z ∼ 0.2 Z_⊙; Russell & Dopita 1992) is falling on to the LMC disk. More recently, Tsuge et al. (2020) investigated the spatial correlation between the intermediate-velocity component and massive stars in the entire LMC field, and suggest that the galactic interaction may have induced a large part (∼70%) of the massive star formation in the LMC. In order to investigate such gas collisions between the gas in the LMC and an inflow gas from the SMC, it is important to study the dust/gas ratio and three-dimensional (3D) dust geometry over the whole LMC.

Furuta et al. (2019) constructed a near-infrared (NIR) dust extinction (A_V) map of the LMC H i ridge region to evaluate the dust/gas ratio. They proposed a method to decompose A_V into different velocity components from the comparison of A_V with hydrogen column densities of different velocities. They decomposed A_V into the three velocity components mentioned above, and found that the dust/gas ratio of the low-velocity component is significantly lower than that of the disk-velocity one. Furthermore, Furuta et al. (2021) developed a new method to evaluate 3D dust geometry from the NIR dust extinction, and applied it to the H i ridge region. They decomposed the resultant 3D A_V map into the three velocity components using the fitting method described in Furuta et al. (2019), and obtained the detailed dust geometry of the different velocity components, from which they found that the gas collision with the inflow gas from the SMC had occurred in 30 Dor prior to N159.

In the present paper, we extend the analysis of the LMC H i ridge region by Furuta et al. (2021) to the entire LMC field. We especially focus on the massive star-forming regions of N44, N79, and N11 for which the tidal interaction between the Magellanic Clouds is reported (Tsuge et al. 2019, 2020). This paper is organized as follows: we summarize the data in section 2 and give a brief review of the method to evaluate the 3D dust extinction in section 3. The resultant dust extinction maps and a comparison with gas observations are presented in section 4. We discuss the 3D geometry and origin of massive star-forming regions in section 5, and summarize our results in section 6.

2 The data

2.1 Point source catalog observed with IRSF/SIRIUS

We used the NIR (J, H, and K_s bands) Magellanic Cloud point source catalog observed with the InfraRed Survey Facility (IRSF) 1.4 m telescope (Kato et al. 2007; hereafter the IRSF catalog). The IRSF catalog covers a 40 ${\mathrm{d}\mathrm{e}\mathrm{g}}^2$ area of the LMC. The 10σ limiting magnitudes of the J, H, and K_s bands are 18.8, 17.8, and 16.6 mag, respectively, which are three magnitudes deeper than those of the 2MASS catalog. In the IRSF catalog, systematic photometric errors due to inappropriate flat-field correction were reported, and thus Furuta et al. (2019) updated the IRSF catalog by applying the corrected flat-field images to the raw data of the catalog (see section 2 of Furuta et al. 2019). In this study, data are selected from the updated IRSF catalog that satisfy the following criteria: (1) the J-, H-, and K_s-band magnitudes are brighter than the 10σ limiting magnitudes of the original IRSF catalog, (2) the shapes of sources are “point like” (the “quality flag” is 1) in all the bands, and (3) at least eight dithered frames are combined in all the bands.

Since Galactic foreground stars leading to underestimation of the dust extinction are included in the IRSF catalog, we identify stellar populations using the classification by Furuta et al. (2019). Stellar populations are classified into three categories that primarily consist of Galactic foreground contamination, red giant branch (RGB) stars, and dwarf stars, using a color–magnitude (J − K vs. K) diagram (figure 1a). In the present paper, only the objects classified as RGB stars are used, a number density map of which is shown in figure 2.

In principle, intrinsically red stars such as young stellar objects (YSOs) can contaminate the selected RGB samples. To check the contamination of YSOs, we match the RGB samples with the YSO candidates in the whole LMC derived from Whitney et al. (2008), Gruendl and Chu (2009), and Seale et al. (2014). We also use the list of YSOs presented by Carlson et al. (2012), who identify YSOs in the nine star-forming regions in the LMC. The total number of YSO candidates is 5714. A matching distance of $5^{\prime \prime }$ is adopted, which is the same value as used by Seale et al. (2014). As a result of the matching, only ∼0.13% (2033/1522279) of the RGB samples in the whole LMC are identified as YSOs. The YSOs matched to our RGB samples are plotted as black points on a color–color (H − K and J − H) diagram in figure 1b. The fraction of contamination of YSOs increases with the dust extinction (A_V); the fraction is $0.31\mathrm{\%}$ (877/278612) for samples with A_V > 1.0 mag, while it is $1.12\mathrm{\%}$ (549/49008) for samples with A_V > 2.0 mag. In this way, a small fraction of YSOs contaminates the RGB samples, leading to overestimation of A_V, especially in star-forming regions. We eliminate such intrinsically red stars from the RGB samples using the percentile method for deriving A_V maps, as described in subsection 3.1.

2.2 Gas tracer

For comparison of the dust extinction with the hydrogen column density, we use the H i data observed with the Australia Telescope Compact Array (ATCA) and Parkes (Kim et al. 2003), and the ¹²CO (J = 1–0) data observed with the NANTEN 4 $\mathrm{m}$ telescope (Fukui et al. 1999). Fukui et al. (2017) and Tsuge et al. (2019) calculate the velocity relative to the galactic rotation defined as V_offset, and decompose the H i and CO integrated intensity maps into three velocity components defined as the L- (low), I- (intermediate), and D- (disk) components according to the integrated velocity ranges of V_offset = −100 to −30, −30 to −10, and −10 to 10 $\mathrm{k}\mathrm{m}{\mathrm{s}}^{-1}$⁠, respectively (for the analysis and an example of the H i spectra, see Tsuge et al. 2019).

2.3 Dust emission

To compare the dust extinction with the dust emission, we use the dust optical depth at 353 GHz (τ₃₅₃) derived from the Planck and IRAS data (Planck Collaboration 2014). The comparison between the dust extinction and emission gives an insight into the dust geometry because the dust extinction only traces the dust located in front of stars, while the dust emission traces all the dust existing along the line of sight. As a dust emission map of the LMC, we use the τ₃₅₃ map created by Tsuge et al. (2019), who subtract the Galactic foreground τ₃₅₃ values. The spatial resolution of the τ₃₅₃ map is adjusted to that of the dust extinction map by the same procedure as the gas maps.

3 Method

Furuta et al. (2021) developed a new method to evaluate the 3D dust geometry for study of the LMC H i ridge region. The detailed procedure is described in their paper (see section 3 of Furuta et al. 2021). Here, we introduce a brief review of the method.

3.1 Derivation of dust distribution along the line of sight

Below we show the calculation procedure of dust distribution along the line of sight for a low H i column density region (⁠$<5\times {10}^{20}\mathrm{c}{\mathrm{m}}^{-2}$⁠) at (α, δ)_J2000.0 = (⁠$5^h{32}^m$⁠, −$69{}^{\circ }{14}^{\prime }$⁠) where we expect to obtain the distribution in the case of no appreciable dust. In figure 3a, we present a color–color diagram (CC diagram) in a spatial bin of $5^{\prime }\times 5^{\prime }$ centered at the region. We first estimate A_V for individual stars from the color excess between the observed and intrinsic colors of RGB stars along the reddening vector (black arrow in figure 3a) on the CC diagram. Here, we assume the intrinsic colors derived from Bessell and Brett (1988) and the reddening vector of Cardelli, Clayton, and Mathis (1989).

Here, it is noted that recent studies show the dependence of RGB colors on stellar metallicity (e.g., Valenti et al. 2004) together with the metallicity gradient in the LMC (e.g., Choudhury et al. 2021), which can lead to an unintended bias in A_V at different galactocentric distances across the LMC with the intrinsic colors of Bessell and Brett (1988) that do not consider the metallicity dependence. Therefore, we check how much the intrinsic colors are affected by the metallicity gradient. Choudhury et al. (2021) find the average metallicity of [Fe/H] = −0.42 dex and a shallow metallicity gradient of −0.008 dex $\mathrm{k}\mathrm{p}{\mathrm{c}}^{-1}$ from the galactic center in the LMC using the color–magnitude diagram of RGB stars. Our A_V map covers radii up to 2.6 kpc from the center, and thus the stellar metallicity can vary by at most −0.02 dex. By changing the stellar metallicity from −0.42 to −0.44 dex considering the above metallicity gradient, the M0-type RGB colors that depend on stellar metallicity proposed by Valenti, Ferraro, and Origlia (2004) change from 0.6922 to 0.6892 mag for J − H and from 0.1059 to 0.1055 mag for H − K. The changes in the colors are much smaller than the average photometric errors of 0.03 and 0.06 mag for the J − H and H − K colors of our samples, respectively, and thus the metallicity gradient in the LMC would not change our results significantly.

In the second step, we sort the estimated A_V from low to high for each spatial bin. The sorted A_V distribution is shown by the red histogram in figure 3b, while the sorted A_V values at every five stars are shown by the red circles as a function of the cumulative number of stars in figure 3c. The A_V values in figure 3c monotonically increase due to the dust existing along the line of sight. However, even when no dust exists along the line of sight, the sorted A_V values increase due to the A_V scatter caused by the photometric errors. Thus, in the third step, we simulate the A_V scatter by a Monte Carlo simulation to evaluate this effect.

We generate a random set of the J − H and H − K colors for individual stars on the CC diagram in each spatial bin under the assumption that the stellar colors follow a 2D Gaussian distribution. Here, we adopt the observed colors and photometric errors as the center and the standard deviation of the Gaussian distribution, respectively. A_V for each simulated star is calculated by the same procedure as the first step. The blue histogram and squares in figures 3b and 3c, respectively, show the resultant distribution of the simulated values, which is considered to be A_V scatter due to the photometric errors alone.

In the next step, the mean percentile values for both observed and simulated A_V are calculated in the range of the X% to (X + 10)% percentile for X = 10% to 80% in steps of 10%. Here, we adopt not $100\mathrm{\%}$ but $90\mathrm{\%}$ as the upper percentile limit for calculation of A_V to avoid overestimation of A_V due to the contamination of intrinsically red stars such as YSOs and asymptotic giant branch (AGB) stars as described in Dobashi et al. (2008). Since the fraction of contamination of YSOs is small (∼1%) even in high A_V regions as mentioned in subsection 2.1, the intrinsically red stars should be eliminated from our samples through the percentile method. Red circles and blue squares in figure 3d are the resultant mean observed and simulated percentile A_V for each X%, respectively. In order to evaluate the intrinsic A_V distribution, we finally subtract the simulated percentile A_V values from the observed one at each percentile range (figure 3e), which is defined as A_V(X%). For example, A_V(80%), corresponding to near-total integrated dust extinction, is the mean A_V from the 80% to 90% percentile of observed A_V after subtracting simulated percentile A_V in the same percentile range.

The uncertainty of A_V(X%) is measured by the error propagation, $\delta A_V(X\mathrm{\%})=\sqrt{{\sigma }_{A_{V,X}}^2+{\sigma }_{d,X}^2}$⁠, where ${\sigma }_{A_{V,X}}$ is the uncertainty of percentile A_V, while σ_d,X virtually corresponds to the uncertainty of the distance. ${\sigma }_{A_{V,X}}$ is calculated by propagation of the A_V uncertainty included in the X% to (X + 10)% percentile. On the other hand, σ_d,X is calculated based on the simple count uncertainty (i.e., $\sqrt{N}$⁠; N is the number of the total stars contained in the bin), and determined by half the difference in A_V between the index of the integer part of (⁠$i_{\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}+\sqrt{N}$⁠) and that of (⁠$i_{\mathrm{c}\mathrm{e}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}-\sqrt{N}$⁠), where i_center is the median index of stars in the range of the X% to (X + 10)% percentile. The uncertainty of $\delta A_V(X\mathrm{\%})$ is used for the comparison of the dust extinction with the hydrogen column density to evaluate the dust geometry and the dust/gas ratio (subsection 3.2).

A_V(X%) in figure 3e is almost constant for all X%, which indicates that dust clouds associated with the LMC are not detected significantly at this spatial bin, as expected from the low column density region. The constant A_V of 0.2 mag is likely to be caused by the Galactic foreground extinction and is actually consistent with the foreground extinction ∼0.2 mag (Dobashi et al. 2008).

In figure 4, we present another result for the spatial bin centered at (α, δ)_J2000.0 = (⁠$5^{h}23m$⁠, −${68}^{\circ }{03}^{\prime }$⁠) in N44 where the D-component is detected (the position is plotted in figure 11). The observed histogram in figure 4a has a peak around A_V = 0 mag and shows A_V systematically larger (>1.5 mag) than the simulated histogram. $A_V(X\mathrm{\%})$ from X = 10% to 30% in figure 4c has an almost constant value of ∼0.2 mag corresponding to the Galactic foreground extinction. On the other hand, $A_V(X\mathrm{\%})$ rapidly increases from $X=40\mathrm{\%}$ onward. These results indicate that a dust cloud is present from $X=40\mathrm{\%}$ to 80%, which is consistent with the existence of the high column density gas of the D-component.

3.2 Decomposition of dust extinction into different velocities

4 Result

4.1 Cumulative dust extinction map over the LMC

In figure 5, we show the A_V(80%) map (i.e., near-total integrated dust extinction map) of the entire LMC field. The spatial resolution of the map is $5^{\prime }$ with a grid size of $1{}_{.}^{\prime }6$⁠. The uncertainty of the map [$\delta A_V(80\mathrm{\%})$] is shown in figure 6. The uncertainty is relatively large near the edge of the map due to low number densities of stars. In these maps, we mask the regions where we see failure in the read-out of the data or inappropriate zero point magnitude calibration.

We compare the A_V(80%) map with the dust extinction map created by Dobashi et al. (2008) using the 2MASS catalog in figure 7. The spatial distribution of high A_V regions such as the H i ridge region and N44 is consistent between the A_V maps in figures 5 and 7. The A_V(80%) map significantly detects dust clouds in N79 and around (α, δ)_J2000.0 = (⁠$5^h{25}^m$⁠, −${66}^{\circ }{15}^{\prime }$⁠), which are not detected in the map of Dobashi et al. (2008). This is probably owing to the fact that the limiting magnitude of the IRSF catalog is three magnitudes deeper than that of the 2MASS catalog. We compare the A_V map of this study and that of Dobashi et al. (2008) on a pixel-by-pixel basis and find a small but systematic difference in A_V (0.3 mag), which can be seen especially in low hydrogen column density regions in figures 5 and 7. This is probably due to the difference in the method of estimating A_V; we select RGB stars and use several intrinsic colors based on the spectral types of RGB stars, whereas Dobashi et al. (2008) did not classify stellar populations and used an intrinsic color averaged for stars of any spectral types and stellar populations in their reference field (i.e., extinction-free region). In the latter case, the A_V values can vary depending on the spectral types and populations of the stars included in each observed field, and thus it is difficult to detect diffuse dust clouds. On the other hand, our dust extinction map is considered to detect diffuse dust clouds because the dust/gas ratio [A_V/N(H)] of the D-component in our study is consistent with that of the LMC estimated from the previous studies as discussed in subsection 5.2.

In figure 5, we recognize that the dust extinction correlates well with the total hydrogen column density. Relatively high A_V regions (A_V > 1.0 mag) can be seen in the massive star-forming regions of the H i ridge, N44, N79, and N11. On the other hand, in the region where Tsuge et al. (2020) identify the diffuse L-component (hereafter the diffuse L-com region), no dust cloud is detected.

4.2 Cumulative dust extinction toward individual star-forming regions

In this section, to evaluate the 3D dust geometry and the dust/gas ratio of different velocity components in the massive star-forming regions of N44, N79, and N11, we investigate spatial correlation between A_V and N(H) of different velocities. Figures 8–10 show the A_V(10%) to A_V(80%) maps of N44, N79, and N11, respectively. Since these maps show the dust extinction integrated from observers to X%, the dust extinction should increase with X%. We also present a comparison of A_V(80%) with N(H) of different velocity components and dust emission of τ₃₅₃ for N44, N79, and N11 in figures 11–13, respectively.

The analysis area of each star-forming region shown in figures 8–10 is determined based on the areas for which Tsuge et al. (2019, 2020) investigate the spatial and velocity distributions of the H i clouds. Although they are larger than the H ii regions of N44, N79, and N11 defined in the Henize (1956) catalog, we adopt these analysis areas covering the L- and/or I-components existing around each

star-forming region to investigate the 3D dust geometry of different velocity components.

4.2.1 N44

From figure 11, we recognize that the dust extinction exhibits a good spatial correlation with N(H) of the D-component and τ₃₅₃, which indicates that all the dust clouds in N44 are located in the LMC disk. On the other hand, we cannot see a clear spatial correlation between the dust extinction and the L- and I-components.

In order to statistically determine whether the A_V counterparts of the L- and/or I-components are detected in the A_V(80%) map, we apply the fitting with equation (2) to the A_V(80%) map. Since N(H) of the L-component is low for a large area as shown in figure 11a, to improve the statistics, we newly define the L+I-component whose velocity range is −100 to −10 $\mathrm{k}\mathrm{m}{\mathrm{s}}^{-1}$ covering the velocity ranges of both the L- and I-components, and decompose A_V into the L+I- and D-components. We first perform the fitting with the D-component alone [i.e., k = D in equation (2)]. In the next step, we add the L+I-components in the fitting (i.e., k = D and L+I). Here, the same X_CO factor is assumed for the D- and L+I-components [i.e., x_D = x_L+I in equation (2)] because the CO emission of the L- and I-components (shown by white contours in figure 11) is detected in only a few areas and thus it is difficult to determine X_CO of the L+I-component by the fitting. To judge whether the fitting is improved significantly, we perform an F-test with a confidence level of 95% corresponding to the F-test probability smaller than 0.05. As a result of the fitting, the ratio of χ² to the degree of freedom (χ²$/$dof) is decreased from 84.51/166 (k = D) to 76.67$/$165 (k = D and L+I); the F-test probability is 6 × 10⁻⁵, indicating that the fitting is significantly improved by adding the L+I-component, and thus the L+I-component is significantly detected in the A_V(80%) map.

To check the validity of introducing the L+I-component, we separate the L+I-component into the L- and I-components and perform the fitting with equation (2) using k = D, L, and I with the same X_CO factor for the D-, L-, and I-components (i.e., x_D = x_L = x_I). As a result, χ²$/$dof decreases from 76.67$/$165 (k = D and L+I) to 75.46$/$164 (k = D, L, and I); the F-test probability is 0.11, indicating that separating the L+I-component does not improve the fitting significantly, and thus it is reasonable to introduce the L+I-component.

To evaluate the 3D geometry of N44, we apply the fitting with equation (2) using the D- and L+I-components to each A_V(X%) map for X = 10% to 80%. Table 1 shows the best-fitting parameters, while figure 14a shows A_V/N(H) of the L+I- and D-components thus obtained as a function of X%. The uncertainties of the parameters are calculated by formal regression errors using the A_V uncertainties of $\delta A_V(X\mathrm{\%})$⁠.

In table 1, many of the derived X_CO factors of N44 for X = 10%–80% are not estimated significantly (i.e., X_CO < 1σ, where σ is the uncertainty of X_CO). This result indicates that A_V counterparts of CO molecules are not detected in many of our A_V maps. Actually, in the total integrated A_V map [A_V(80%)], the X_CO factor is estimated significantly (X_CO > 1σ). From table 1, we find that the A_V/N(H) ratio at X = 80% for the L+I-component of N44 is lower than that for the D-component. Since X_CO is known to depend on the metallicty (Bolatto et al. 2013), we need to check the validity of using the same X_CO factor for the L+I and D-components in the fitting. Hence, by tentatively setting X_CO for the L+I-component as a new free parameter independent of the D-component, χ²$/$dof is decreased from 76.67$/$165 (k = D, L+I with x_D = x_L+I) to 76.66/164 (k = D, L+I with x_D ≠ x_L+I); the F-test probability is 0.86, which means that the fitting is not improved significantly and using the same X_CO for the L+I- and D-components does not affect our results. This is probably due to the fact that the CO emission of the L+I-component is detected in only small areas and thus it is difficult to detect the difference in X_CO that is expected to be caused by the metallicity difference between the L+I- and D-components.

4.2.2 N79

In figure 12b, the D-component gas correlates well with the dust extinction map, although the right-hand side of the map is noisy due to the low number densities of the stars (figure 2). Dust emission τ₃₅₃ also shows the spatial correlation with the dust extinction, indicating that all the dust is present in the LMC disk.

To check whether the A_V counterparts of the I-component are detected significantly, we apply the same fitting procedure for N44 to the A_V(80%) map of N79. By adding the L+I-component to the fitting with the D-component alone, χ²$/$dof is improved from 195.28/231 (k = D) to 163.66/230 (k = D and L+I). The F-test probability is 2 × 10⁻¹⁰, meaning that the fitting is significantly improved, and thus the A_V(80%) map calls for the presence of the L+I-component. The parameters estimated from the fitting with the D- and L+I-components are summarized in table 1 and plotted in figure 14b.

From table 1, we find that X_CO for N79 is not detected significantly at $X=10\mathrm{\%}$–40%, while it is detected with significance higher than 1σ at $X=50\mathrm{\%}$–80%, which indicates that A_V counterparts of CO molecules are detected only at large $X\mathrm{\%}$⁠. A_V/N(H) at $X=80\mathrm{\%}$ for the L+I-component of N79 is consistent with that for the D-component, and thus using the same X_CO for the L+I- and D-components in the fitting is considered to be reasonable.

4.2.3 N11

From figures 13a and 13b, we recognize that dust clouds are detected in the regions where the D- and I-components are detected. In fact, by adding the L+I-component to the fitting with the D-component alone, χ²/dof is improved from 37.07$/$81 to 33.93$/$80; the F-test probability is 8 × 10⁻³ and again the A_V(80%) map requires the presence of the L+I-component. In figure 13c, the dust extinction spatially correlates well with τ₃₅₃, which indicates that all the dust exists in the LMC disk. We perform the same fitting to the A_V(X%) maps of N11 as to those of N44. The derived fitting parameters are shown in table 1 and plotted in figure 14c.

From table 1, we recognize that the A_V/N(H) ratio at $X=80\mathrm{\%}$ for the L+I-component of N11 is consistent with that for the D-component within the errors, and thus using the same X_CO factor for the L+I- and D-components is considered to be reasonable. All of the X_CO values for $X=10\mathrm{\%}$–80% are lower than 1σ, which indicates that the A_V counterparts of the CO molecules are not detected even in the A_V(80%) map, or X_CO cannot be estimated precisely by the fitting due to the large A_V uncertainties.

4.3 Dust geometry for the diffuse L-com region

In the “diffuse L-com region” that is located in the west of N44 and marked in figure 5, Tsuge et al. (2020) find that there is no significant H ii region in the area where the L-component exists. They also find that there is no signature of deceleration of the L-component in the first moment map of the diffuse L-com region, and thus they suggest that the L-component in this region is located behind the LMC disk, and the gas collision is yet to occur. To verify this hypothesis, we evaluate the 3D geometry of this region.

In figure 15, we show the A_V(80%) maps with N(H) of the L-, I-, and D-components and τ₃₅₃. In figure 15a, we cannot see a clear spatial correlation between the L-component and the dust extinction. The D-component extends in this region and roughly correlates with the dust extinction in figure 15c, while the CO emission of the D-component (shown by white contours in figure 15c) does not correlate with the dust extinction, especially in the south of the map, which indicates that the molecular gas is located on the far side of the LMC disk.

To determine whether the A_V counterparts of the L- and/or I-components are detected significantly, we apply the linear regression fitting to the A_V(80%) map similarly to N44. Since the A_V counterparts of molecular gas are not detected as mentioned above, we mask the regions in the A_V map where the CO intensity is higher than 1.5σ and perform the fitting with equation (2) using only the H i data. By adding the L+I-component to the fitting with the D-component alone, χ²$/$dof is improved from 248.72/210 (k = D) to 230.11/209 (k = D and L+I); the F-test probability is 6 × 10⁻⁵, meaning that the A_V(80%) map calls for the presence of the L+I-component. In addition, as in the study of N44, we separate the L+I-component into the L- and I-components and perform the fitting using k = D, L, and I. As a result, χ²$/$dof is decreased from 230.11/209 to 223.49/208. The F-test probability is 0.014, which means that the fitting is significantly improved by separating the L+I-component. The estimated parameters of A_V/N(H) of the L-, I- and D-components for the A_V(80%) map are $(0.37\pm 0.40,2.63\pm 0.67,\text{and}2.70\times 0.38)\times {10}^{-22}\mathrm{mag}\mathrm{c}{\mathrm{m}}^2$⁠, respectively. A_V/N(H) of the L-component is not detected significantly, which indicates that the L-component is located behind the LMC disk and does not contribute to the dust extinction. On the other hand, the I-component is significantly detected and contributes to the dust extinction. Such a difference between the L- and I-components results in an improvement in the fitting by separating the L+I-component.

To evaluate the geometry of the I- and D-components, we apply the fitting with equation (2) using N(H i) of the I- and D-components to each A_V(X%) map. The estimated parameters of A_V/N(H) are summarized in table 2 and shown in figure 16. A_V/N(H) of the D-component monotonically increases with X%, indicating that the dust of the D-component extends along the line of sight. The A_V/N(H) ratio of the I-component agrees with that of the D-component. Thus, the dust of the I-component is likely to be distributed similarly to that of the D-component. In summary, in the diffuse L-com region, the L-component is located behind the LMC disk, while the I- and D-components extend along the line of sight, which is consistent with the hypothesis that gas collision is yet to occur, as mentioned by Tsuge et al. (2020).

5 Discussion

5.1 Dust geometry of individual star-forming regions

In this section, we evaluate the dust geometry of N44, N79, and N11 from the estimated fitting parameter of a_k [i.e., A_V/N(H)] of each X%. We first consider the dust geometry of N44. The likely 3D dust geometry of the D- and L+I-components for N44 is illustrated in figure 17a. First, A_V/N(H) of the D-component in figure 14a monotonically increases with X%, and thus the D-component is expected to extend along the line of sight as shown by a red rectangle in figure 17a, which is consistent with the general idea that the gas of the LMC disk is mixed with stars. A_V/N(H) of the L+I-component is significantly detected from X = 30% onward, which indicates that the head of the L+I-component is located at X = 30%. In addition, A_V/N(H) of the L+I-component is almost constant from X = 60% to 80%, indicating that there is little dust of the L+I-component beyond X = 60%. As a whole, the L+I-component in N44 is likely to extend from X = 30% to 60% as shown by a blue rectangle in figure 17a. The dust geometry suggests that the L+I-component is penetrating the D-component. This trend supports the idea that gas collision between the gas of the LMC disk and an inflow gas as seen in the L+I-component may have induced the massive star formation in N44 (Tsuge et al. 2019).

Similarly to N44, we discuss the 3D dust geometry of N79 and N11. In both regions, from the monotonic increase of A_V/N(H) of the D-component with X% in figures 14b and 14c, the D-component is expected to extend along the line of sight as shown by red rectangles in figures 17b and 17c. A_V/N(H) of the L+I-component is not detected significantly at X = 10%–20% and X =10%–30% for N79 and N11, respectively (see table 1), and thus the L+I-component is expected to extend beyond X = 30% for N79 and beyond X = 40% for N11 as shown by blue rectangles in figures 17b and 17c, respectively.

We compare the dust geometry of N44, N79, and N11 as well as the H i ridge region with the evolutionary stages of giant molecular clouds (GMCs) proposed by Kawamura et al. (2009). We summarize the positions of the L- or L+I-components and evolutionary stages of the relevant GMCs in table 3. The H i ridge region is separated into three regions based on the geometry of the L-component by Furuta et al. (2021). They define them as regions 1–3 from north to south; regions 1 and 2 contain 30 Dor and N159, respectively (see figure 14 in Furuta et al. 2021). The evolutionary stages are classified into four stages, Types I–III and the last stage in order from youngest to oldest. From the positions relative to the LMC disk of the L+I-component for N44, N79, and N11 in table 3, we recognize that the gas collision in these regions occurred later than in 30 Dor but occurred earlier than in region 3, which is consistent with the evolutionary sequences of the GMCs. The evolutionary stage of GMCs in N79 is earlier than those of N159, N44, and N11. However, we cannot see clear differences in the timing of gas collisions between these regions. This is likely to be caused by the large A_V uncertainties, especially for N79 and N11, due to the low number densities of the stars.

In order to discuss the relationship between the timing of the gas collision and the evolutionary stages of GMCs quantitatively, we compare the crossing timescale of the gas collision with the transition timescale of the evolutionary stages. For example, from the comparison of the dust geometry of 30 Dor with that of N44, the head of the L+I-component is separated by the distance of the 30% percentile. Considering am LMC disk thickness of 2 kpc (Balbinot et al. 2015), this corresponds to 0.6 kpc. Assuming the velocity of the inflow gas (L-component) to be ∼100 km s⁻¹ (Tsuge et al. 2020), it takes 6 Myr for the inflow gas to cross 0.6 kpc, which is consistent with the timescale of 7 Myr for the transition from Type III GMCs in N44 to the last-stage GMCs in 30 Dor (Kawamura et al. 2009). Therefore, our result can reasonably explain the difference in the evolutionary stages of the GMCs across the LMC.

5.2 Origin of massive star-forming regions

We discuss the origins of the massive star-forming regions of N44, N79, and N11 from the A_V/N(H) ratio at $X=80\mathrm{\%}$⁠. We summarize A_V/N(H) for these regions as well as the H i ridge region in table 4. Furuta et al. (2021) find a difference of a factor of two in A_V/N(H) between the D- and other components for the H i ridge region (see table 4). From the existence of the low-metallicity gas, they propose that the massive star formation in 30 Dor was triggered by galactic interaction between the LMC and the SMC.

The A_V/N(H) ratio of the D-component at X =80% for the star-forming regions of N44, N79, and N11 is similar to that of the H i ridge region (see table 4). The resultant A_V/N(H) ratio is nearly half of that in the Milky Way $(5.34\times {10}^{-22}\mathrm{mag}\mathrm{c}{\mathrm{m}}^2$ for R_V = 3.1; Bohlin et al. 1978), which is consistent with the dust/gas ratio for the LMC of ∼1/2–1/3 of the Galactic value (Leroy et al. 2007; Roman-Duval et al. 2014). The X_CO factors of the D-component at $X=80\mathrm{\%}$ are consistent within the errors between all the star-forming regions shown in table 4, although the uncertainties are large, especially in N11. The resultant X_CO factors for all the star-forming regions are similar to that of the LMC estimated from CO and [C ii] observations after correcting for the photodissociation of CO molecules due to UV radiation [$X_{\mathrm{C}\mathrm{O}}\sim 2.9\times {10}^{20}\mathrm{c}{\mathrm{m}}^{-2}/(\mathrm{K}\mathrm{k}\mathrm{m}{\mathrm{s}}^{-1})$⁠; Pineda et al. 2017].

On the other hand, A_V/N(H) of the L+I-component significantly varies between the star-forming regions. For N44, A_V/N(H) of the L+I-component is ∼1/5 of the Galactic value, which is similar to the dust/gas ratio for the SMC of ∼1/6 of the Galactic value (Roman-Duval et al. 2014). The existence of low-metallicity gas in the L+I-component in N44 supports the contamination of the inflow gas from the SMC, as proposed by Tsuge et al. (2019).

In N44, expansion of H i shell is reported (Kim et al. 1998). Around the shell, three episodes of massive star formation are found; one is the ∼10 Myr-old star formation inside the shell, another is the ∼5 Myr-old star formation on the shell rims, and the other is the YSOs (typically <1 Myr-old) (e.g., Oey & Massey 1995; Chen et al. 2009; Carlson et al. 2012). Tsuge et al. (2019) propose that the 5 Myr-old massive star formation was triggered by the galactic interaction between the LMC and the SMC as an alternative scenario of star formation triggered by the expansion of the H i shell from the existence of low-metallicity gas and signature of gas collision between the L- and I-components in addition to the lack of energy supply from the shell to explain the motions of the L-, I-, and D-components in N44. The age of the stellar population of N44 (∼5 Myr) is younger than that around the 30 Dor region of 8.1 Myr (Schneider et al. 2018), which agrees with the difference in the evolutionary stages of GMCs between N44 and 30 Dor (see table 3). Therefore, from the low A_V/N(H) ratio of the L-component and the gas-colliding geometry in figure 17a, we suggest that the massive star formation around the shell rims in N44 was triggered by the galactic interaction between the Magellanic Clouds, similarly to the H i ridge region.

For N79 and N11, A_V/N(H) of the L+I-component at X = 80% is nearly half of the Galactic value, which is similar to that of the D-component. According to the recent numerical simulations of the tidal interaction between the LMC and SMC (Tsuge et al. 2020), the gas currently falling on to the LMC disk as the L+I-component consists of gas from the LMC as well as from the SMC. We thus expect that in some places the metallicity will not be very different from that of the LMC, if the L+I-component is dominated locally by the LMC gas. In addition, an analysis of the dust/gas ratio based on the Planck dust emission of τ₃₅₃ (K. Tsuge et al. in preparation ) indicates that the dust/gas ratio toward N79 and N11 is about two times larger than that in the H i ridge region and N44. This is consistent with the present results. It is therefore possible that the N79 and N11 regions were triggered by the L+I-component that originated in the LMC. We thus suggest that the trigger in the two regions was due to internal interaction, although the possibility of tidal interaction cannot be ruled out because the numerical simulation by Tsuge et al. (2020) suggests that part of the LMC gas is stripped off from the LMC disk in the interaction and then merges with the SMC gas to fall down to the LMC disk.

N79 is located at the intersection of the galactic bar-end with the spiral arm. Ochsendorf et al. (2017) suggest that the unique location may provide the gas accumulation and compression to create massive star formations. A similar case is reported for W 43, which is located at the intersection of the Galactic bar-end with the Scutum Arm. Kohno et al. (2021) find several velocity components with a velocity difference of ∼20 km ${\mathrm{s}}^{-1}$ in W 43 and suggest that gas collision between them triggered the starburst. The velocity difference is similar to that between the D- and I-components in our study. Therefore, we suggest that the massive star formation in N79 was triggered by the collision of gas in the stellar bar with inflow gas from the spiral arm.

In N11, several OB associations are located at the periphery of a central cavity with a diameter of 170 pc evacuated by the rich OB association of LH9 (Lucke & Hodge 1970). Previous studies suggest that the OB association of LH10 located in the peripheral clouds was triggered by the expansion of the supershell blown by LH9 (e.g., Barbá et al. 2003; Hatano et al. 2006; Celis Peña et al. 2019). The expansion velocity of the supershell is estimated to be $\sim 20\mathrm{k}\mathrm{m}{\mathrm{s}}^{-1}$ (Meaburn et al. 1989), which agrees with the difference in the velocity between the I- and D-components. The spatial distribution of the I-component is similar to the supershell structure (Tsuge et al. 2020). Using an I-component velocity of 30 km s⁻¹, it takes 2.8 Myr for the I-component to travel the radius of the supershell (85 pc), which is similar to the 2 Myr age difference between LH9 and LH10 (Walborn & Parker 1992). Therefore, our results support the idea that the massive star formation of LH10 in N11 was triggered by internal interaction between the supershell and the surrounding ISM.

In summary, when N(H) of the L-component is detected significantly as in the cases of the H i ridge region and N44, A_V/N(H) of the L+I-component is lower than that of the D-component. It is likely that the L+I-component originates from inflow gas from the SMC and the star formation is triggered by external interaction between the Magellanic Clouds. On the other hand, in the cases that N(H) of the L-component is not detected significantly, as in the cases of N79 and N11, A_V/N(H) of the L+I-component shows fair agreement with that of the D-component. In these regions, internal interactions such as gas converging from the spiral arm and the expansion of a supershell are likely to trigger the massive star formation.

6 Conclusion

We derive a three-dimensional dust extinction map of the massive star-forming regions of N44, N79, and N11 in the entire LMC using the percentile method proposed by Furuta et al. (2021). Our total integrated dust extinction map calls for more abundant dust in the star-forming regions than the previous near-infrared dust extinction map constructed by Dobashi et al. (2008). From a comparison of the dust extinction with the hydrogen column densities of different velocity components, we investigate the three-dimensional geometry and the dust/gas ratio of the different velocity components for each star-forming region. Our main results are as follows:

1. The dust geometry of N44, N79, and N11 suggests that the L+I-component is penetrating the LMC disk. Considering the present results together with the dust geometry of the H i ridge region estimated by Furuta et al. (2021), the difference in the timing of the gas collision agrees with the difference in the evolutionary stages of the giant molecular cloud related to each star-forming region (Kawamura et al. 2009).

2. In the diffuse L-com region, the dust geometry indicates that the L-component is located behind the LMC disk. This geometry is consistent with the hypothesis that the gas collision is yet to occur in this region, as expected from the lack of a signature of deceleration of the L-component and the lack of massive stars (Tsuge et al. 2020).

3. The A_V/N(H) values and the X_CO factors of the D-component at $X=80\mathrm{\%}$ are almost the same for all the star-forming regions, while the A_V/N(H) ratios of the L+I-component vary between them. For N44, A_V/N(H) of the L+I-component is 1/5 of the Galactic value, which is similar to the dust/gas ratio for the SMC (Roman-Duval et al. 2014). From the low A_V/N(H) ratio, we suggest that external interaction between the LMC and the SMC triggered the star formation in N44, similarly to 30 Dor in the H i ridge region.

4. For N79 and N11, the A_V/N(H) ratio of the L+I-component is 1/2 of the Galactic value, which is similar to that of the D-component. Thus, we suggest that the star formation in these regions was triggered by internal interaction although the possibility of the tidal interaction cannot be ruled out. We suggest that the massive star formation of N79 was triggered by a converging gas flow from the spiral arm based on the unique location of N79, while that of N11 was triggered by interaction between the expansion of a supershell and the surrounding ISM.

Acknowledgements

We thank Prof. Kazuhito Dobashi for kindly giving us the data for their A_V map. We also thank the referee for giving us helpful comments. The IRSF project is a collaboration between Nagoya University and the SAAO supported by Grants-in-Aid for Scientific Research on Priority Areas (A) (Nos. 10147207 and 10147214) and the Optical and Near-Infrared Astronomy Inter-University Cooperation Program, from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan and the National Research Foundation (NRF) of South Africa. This research was financially supported by a Grant-in-Aid for JSPS Fellows Grant Number 20J12119.

References