Star-forming Clumps in Local Luminous Infrared Galaxies

The Paβ observations were taken with the HST WFC3 camera (Project ID: 13690; PI: T. Díaz-Santos), which has a plate scale of 012 pixel⁻¹ using the F130N and F132N narrowband filters and the F110W broadband filter. The total integration times were 120 s for each broadband filter and about 1100 s (19 minutes) for each narrowband filter. The F130N filter is used to measure the underlying continuum emission next to the line. The 2' FOV of WFC3-IR allows for galaxies in a pair to be observed simultaneously. Each observation was again divided into three individual exposures with offsets of 5 pixels.

The archival Paα observations were taken with the HST NIC2 camera, which has a plate scale of 0076 pixel⁻¹ and a 192 FOV using the F187N and F190N narrowband filters. This FOV allows for detection of Pα emission up to a radial distance of ∼5 kpc from the galaxy nucleus. From our Paβ data, which cover a much larger FOV, we find that most of the Paschen emission is indeed contained within 5 kpc from the galaxy nucleus. For objects with redshifts between 0.0093 and 0.0174 the Paα emission line, at 1.87 μm, is captured with the F190N filter, while the adjacent filter, F187N, is used for continuum subtraction. The typical total integration times were 900–950 s for each narrowband filter. Each observation was divided into three individual exposures with offsets of 5 pixels. For details on the data reduction of the Paα observations we refer the reader to Alonso-Herrero et al. (2006). Observational details are given for every galaxy in our combined Paα and Paβ sample in Table 1.

In order to have a consistent comparison of star-forming clumps between all the galaxies in the combined sample, we rebinned and smoothed all images to a common physical resolution of 91 pc pixel⁻¹, which corresponds to the resolution of the most distant galaxy in our sample. To create the continuum-subtracted Paα and Paβ line images, the narrowband images (F190N and F187N for Paα and F132N and F130N for Paβ) were convolved with the corresponding line's point-spread function (PSF) to create images with matched pixel scales and PSFs before subtraction. This allowed for a final clean continuum-subtracted line image at a common pixel scale for all galaxies. Figure 1 shows the broadband F110W HST images and the continuum-subtracted Paβ line images for both NGC 6786 and NGC 6090. Depending on the position of the line in the narrowband filter, there can be significant flux loss (Alonso-Herrero et al. 2000). We model a Gaussian emission-line width of 200 km s⁻¹ to estimate and correct for the flux loss due to the position of the line in the narrowband filter. In general, the position of the Paschen line is well contained in the narrowband filter and has less than a 5% flux loss. One exception is NGC 5257-8, which does have significant estimated flux loss of 70% owing to the position of Paβ in the narrowband filter.

Zoom In Zoom Out Reset image size

Our ability to compare star-forming clump properties to high-z and high-resolution simulations depends on being able to consistently detect star-forming clumps with a wide range of luminosities and sizes. In order to detect the clumps and measure their properties, we use the continuum-subtracted line images and the python code astrodendro.¹⁷ Astrodendro computes a dendrogram, which is a branching tree of hierarchical structure, of an image. This procedure compares each possible clump to the local background, successfully identifying clumps down to the lowest detectable flux limit (Figure 2).

Zoom In Zoom Out Reset image size

There are three important parameters in the astrodendro code that affect the identification of clumps. The first is the minimum size of the region. Since the FWHM of the data is ∼1 pixel, we set a minimum size of 3 pixels to be accepted as a clump. The 3-pixel minimum size maximizes the detection of the smallest clumps while avoiding blending of clumps. Then, a minimum threshold above the general background level is also required, which we set at 5σ above the general sky level of the image. This ensures that we have a resolved clump that is well above any background noise or diffuse emission. Lastly, a minimum significance for the substructure is set that determines when a substructure is brighter than the local background. The minimum significance is the difference in surface density required between a structure and any substructures for a substructure to be retained. When a substructure is retained, it is shown as a new branch in the dendrogram tree in Figure 2. We set the minimum significance to 1σ above the local background. We then take the upper branch of the dendrogram tree (red lines in the left panel of Figure 2) as our final clump regions. Astrodendro also detects any bright artifacts left in the images from bright stars or noise. Therefore, the clumps are all confirmed by visual inspection, and any sources not associated with the galaxy are removed. A detailed comparison of astrodendro to the other commonly used clumpfind algorithm (Williams et al. 1994) for identifying and measuring clump properties in our sample LIRGs is given in the Appendix.

Both the continuum emission and the line emission must be accurately measured in order to determine the ages and masses of all the clumps. The same clump regions found with astrodendro on the Paα and Paβ line images were used on the narrowband continuum images. We subtract the local background, determined by the region surrounding the clump and not including any other clumps, to measure the clump fluxes. The local background regions were visually inspected for each clump in line and continuum images in order to avoid hot spots or other continuum features that could cause an oversubtraction of the clump line and/or continuum fluxes. If they were contaminated by continuum features, alternative clean background regions near the source were selected. Local background subtraction is incredibly important to ensure that the resulting SFRs, ages, and masses are for the clump and not contaminated by underlying stellar populations (Guo et al. 2018).

We determine the errors in the clump sizes using a Monte Carlo method. Astrodendro is rerun 1000 times randomly adjusting the flux of each pixel in the image while sampling from a Poisson distribution. The standard deviation of the size is determined from the measured radii of each region from the 1000 iterations. This allows us to characterize the accuracy of the region's size based on the noise in the image.

The ages and masses of the clumps were computed by comparing the measured line and continuum fluxes of the clumps to Starburst99 models (Leitherer et al. 1999). We estimated the Paα and Paβ equivalent widths of every clump and matched them to the corresponding equivalent widths of the model spectra that were calculated in time steps of 0.1 Myr to determine the clump ages. We compared to both an instantaneous burst single stellar population model and a continuous star formation model with a Kroupa initial mass function (IMF; Kroupa 2001) and solar metallicity for our clumps. The masses of the clumps were calculated with the background-subtracted continuum luminosities measured from the narrowband filters (F130N or F187N) and the estimated mass-to-light ratio from the model M(M_⊙) = L_{cont–filter} × (M/L). The mass-to-light ratio varies with the age of the stellar population and was estimated for each clump at the fitted age.

Figure 3 shows the distributions of ages and masses for extranuclear star-forming clumps. The instantaneous burst model gave a narrow range of ages from $2\times {10}^{6}$ yr to 6.8 × 10⁶ yr with a median age of 4 (+0.8, −1) × 10⁶yr. All uncertainties are given at the 68% confidence levels. The continuous burst has a wider range of stellar ages from 1.5 × 10⁶ yr to 1.6 × 10¹⁰ yr, with a median age of 8.7 (+6, −2) × 10⁶ yr. We found that both the instantaneous burst and continuous star formation models give masses of extranuclear clumps that range from 10⁴ to 10⁹ M_⊙ with a similar median clump mass of ∼5 × 10⁵ M_⊙. The median and range of clump ages and masses using both the instantaneous burst and continuous star formation models for the extranuclear clumps are given in Table 2. Since the nuclei can be contaminated by AGNs and may be effected by higher levels of extinction, we exclude the nuclei from Figure 3 and Table 2.

Zoom In Zoom Out Reset image size

Even with careful background subtraction, we were unable to calculate accurate ages and masses for 45 extranuclear clumps. These clumps have large errors on the continuum measurements, which lead to errors in the age estimate that are as large as the age. They have very little continuum detected and therefore large errors that do not allow for accurate estimates of the ages and masses. We therefore exclude these clumps from our discussion of ages, masses, and SFRs. Further discussion of ages, masses, and SFRs of extranuclear clumps will only include the 706 clumps. A further 21 of these extranuclear clumps have unconstrained ages owing to equivalent widths that are outside of the range predicted by the continuous star formation history (SFH) model. These clumps are clearly identified in both line and continuum emission and are therefore included in the future discussion of SFRs.

The clumps we find using the astrodendro technique span a large range of luminosities and sizes as shown in Figure 4. The area of the clump is determined by the number of pixels in the region, and the effective radius is defined as ${\rm{r}}=\sqrt{A/\pi }$. To allow for an equal comparison of both Paα and Paβ images and other Hα clump studies, we convert all clump luminosities to SFRs. We use a conversion factor determined from Starburst99 stellar population models assuming a continuous star formation, Kroupa IMF, and stellar metallicity at the appropriate age for each clump. This conversion factor uses the "case B" line intensity ratios and assumes no extinction. While extinction from dust obscuration is much less in the near-infrared than in the optical for U/LIRGs, it can still affect the nuclei of the galaxies and decreases quickly in the extranuclear regions (Díaz-Santos et al. 2011). Previous studies using VLT/SINFONI data have shown that local (z < 0.02) LIRGs have subkiloparsec clumpy dust structures. These regions have a wide range of A_v from 1 to 20 mag and an average A_v = 5.3 mag within their 3 × 3 kpc FOV (Piqueras López et al. 2013). We therefore cannot apply a single galactic A_v to our star-forming clumps since the dust obscuration is extremely patchy and each galaxy presents different spatial clump distributions. In a study of star clusters in GOALS galaxies, Linden et al. (2019) estimated the extinction to individual extranuclear clusters using far-UV (FUV), B, and I photometry. Thirty-two of our star-forming clumps distributed over six galaxies overlap with their sample of detected extranuclear star clusters. The estimated A_v for these extranuclear regions ranges from  0.2 to 1.8 mag with an average A_v = 1, which implies an A_NIR = 0.1–0.2 mag. Therefore, while extinction corrections can be important for some extremely dusty clumps, the extinction is highly spatially variable and can be fairly low outside of the circumnuclear regions. Our nuclear clumps likely have the largest corrections, but these are unknown. The fluxes and derived SFRs of the nuclear clumps are therefore, effectively, lower limits.

Zoom In Zoom Out Reset image size

Due to large continuum measurements and accordingly low equivalent width in some of the nuclei, 20 of the 59 nuclear clumps have equivalent widths that are below the range predicted by the continuous SFH models. For the 20 nuclei and the 21 extranuclear clumps with equivalent widths not covered in the continuous SFH models, we cannot determine an age. The clump SFRs are therefore calculated using the asymptotic calibration value from the Starburst99 continuous SF model at 10⁸ yr (SFR/L_Hα = 5.37 × 10⁻⁴² [M_⊙ yr⁻¹]/[erg s⁻¹]). While the Starburst99 continuous SF model provides a variable luminosity to SFR calibration with age, the calibration starts to asymptote to a constant value at an age of 5 × 10⁶ yr and reaches a constant value by 10⁸ yr. The asymptotic calibration reached for our Starburst99 models is the same calibration determined by Murphy et al. (2011).

In Figures 4 and 5, we see that the clump SFRs span a wide range from 2 × 10⁻³ to 4.4 M_⊙ yr⁻¹ with nuclear clumps that have SFRs of 5 × 10⁻³ to 9.6 M_⊙ yr⁻¹. We find clump radii that range from our minimum resolvable size of 89 pc up to 883 pc, with extranuclear clumps reaching sizes of 678 pc and a median of 171 (+47, −26) pc.

Zoom In Zoom Out Reset image size

We provide the measurements of all clumps with determined SFRs in our Paα and Paβ sample in Table 3. The clumps are identified by their host galaxy and ID number (GalaxyName_c#). Data for nuclei in each galaxy are also given and identified by the host galaxy name and nuclei number (GalaxyName_n#). The estimated age and mass of each clump are given, assuming a continuous SFH and solar metallicity, as well as the clump SFR and effective radius.

The SFR surface density (Σ_SFR) of the extranuclear GOALS clumps ranges from 0.045 to 14.86 M_⊙ yr⁻¹ kpc⁻². Star-forming regions with Σ_SFR > 1 M_⊙ yr⁻¹ kpc⁻² are often referred to as "starbursting" since extranuclear star-forming regions in normal spiral galaxies are found to have Σ_SFR below this cutoff (Maragkoudakis et al. 2017; Nguyen-Luong et al. 2016; Alonso-Herrero et al. 2000; Kennicutt 1998). Furthermore, Cosens et al. (2018) have also found a break in the size-luminosity scaling relation at a ${{\rm{\Sigma }}}_{{\rm{S}}{\rm{F}}{\rm{R}}}\,=1\,{M}_{\odot }\,{{\rm{y}}{\rm{r}}}^{-1}\,{{\rm{k}}{\rm{p}}{\rm{c}}}^{-1}$. Figure 6 shows how the majority of the clumps lie below Σ_SFR = 1 M_⊙ yr⁻¹ kpc⁻², with 13% of the extranuclear clumps in the "starburst" region. The median Σ_SFR of extranuclear clumps is 0.31 (+0.19, −0.10) M_⊙ yr⁻¹ kpc⁻², while nuclear clumps have a higher median Σ_SFR= 1.4 (+0.8, −0.5) M_⊙ yr⁻¹ kpc⁻² and are more likely to be starbursting, with 64% of the nuclear clumps in the "starburst" region. The median and range of clump SFRs and Σ_SFR values for the extranuclear clumps are given in Table 2. Piqueras López et al. (2016) observed a sample of 17 LIRGs and ULIRGs with the VLT/SINFONI integral field unit and found a wider range of SFR surface densities for clumps in the range of $0.1-60\,{{\rm{\Sigma }}}_{\mathrm{SFR}}$, although these data were corrected for extinction (from the Brγ and Brδ ratios). We converted the Piqueras López et al. (2016) results from Salpeter to Kroupa IMF to be comparable to the results presented in this paper. The wide range of SFR surface densities found by Piqueras López et al. (2016) is due to their applied extinction correction. The clumps with the largest ${{\rm{\Sigma }}}_{\mathrm{SFR}}$ in their study are the ones with the largest inferred extinction, and these tend to be in the nuclei, making up a small fraction of the distribution. Therefore, the median observed and extinction-corrected ${{\rm{\Sigma }}}_{\mathrm{SFR}}$ for the ULIRGs in the Piqueras López et al. (2016) sample are ${{\rm{\Sigma }}}_{\mathrm{SFR}}^{\mathrm{obs}}=0.11$ ${{\rm{M}}}_{\odot }$ yr⁻¹ kpc⁻² and ${{\rm{\Sigma }}}_{\mathrm{SFR}}^{\mathrm{corr}}=0.16$ ${{\rm{M}}}_{\odot }$ yr⁻¹ kpc⁻², similar to the median extranuclear Σ_SFR found in our sample of Σ_SFR = 0.31.

Zoom In Zoom Out Reset image size

Even though our galaxy sample is selected to be infrared luminous, the majority of our extranuclear star-forming clumps have Σ_SFR at the upper end of Σ_SFR values that are found in normal spiral galaxies, with 13% of the clumps in the starburst region.

The total SFR, derived from the HST data, includes the clumpy and diffuse narrowband emission. For each galaxy, this is determined by converting the Paschen luminosity to Hα luminosity assuming the "case B" recombination and then converting the Hα luminosity to SFR using the calibration from Murphy et al. (2011) (SFR/L_Hα = 5.37 × 10⁻⁴² [M_⊙ yr⁻¹]/[erg s⁻¹]). This calibration assumes a Kroupa IMF with continuous SFR and solar metallicity integrated over a timescale of 100 Myr. We use the Murphy et al. (2011) calibration for obtaining the total SFR of the galaxies since we are averaging over a range of sizes and ages of star-forming clumps and diffuse emission. Figure 7 shows the fraction of the total SFR, as derived from the infrared emission, that is detected in each galaxy in the narrowband HST images. The total infrared SFR of the galaxies is calculated by converting the total infrared luminosity of the galaxies, L_IR, to an SFR (Murphy et al. 2011).

Zoom In Zoom Out Reset image size

There is a large range in the percentage of the total SFR, as traced in the far-infrared (FIR), recovered in our Paschen data (Figure 7). The median ratio of Pβ or Pα to total FIR-derived star formation in our sample is 9%. The galaxy with the lowest detected Paschen emission is IC 0860, with only 0.14% of the total infrared emission detected, and is not shown in Figure 7. This galaxy has no extranuclear clumps, and all detected emission is from the nucleus. This source must be heavily enshrouded even in the near-infrared.

Figure 7 shows the percentage of the Paschen emission for each galaxy observed in clumps. The percentage of the total Paschen emission contained in extranuclear clumps is 0% to 65%, with a median of 14.7%. When the nuclei are included, 10%–100% of the Paschen emission is contained in clumps, with the remaining emission in diffuse emission. The nuclear regions can contain a significant fraction of the detected Paschen emission even without extinction correction. Some galaxies, such as NGC 5258 and NGC 6786, are dominated by extranuclear star formation with no detectable nuclear emission in Paβ, whereas other galaxies, such as IRAS F02437+2122, IRAS 23436+5257, and NGC 7591, are dominated by strong nuclear emission with little or no extranuclear star-forming clumps.

The LF of star-forming clumps in local galaxies can be well approximated by a power law with a slope of α = −2 (e.g., Zhang & Fall 1999; de Grijs et al. 2003; Chandar et al. 2015; Cook et al. 2016). Local LIRGs cover an SFR of 20 < SFR [M_⊙ yr⁻¹] < 200, thus making up the high-LF end.

Previous studies have shown that different binning methods can change the measured slope of the LF (Maíz Apellániz & Úbeda 2005, Cook et al. 2016). While the equal luminosity-sized bins is the most commonly used method, Maíz Apellániz & Úbeda (2005) showed that this caused systematic biases at the brightest luminosity bins owing to low number statistics. Therefore, we chose to use variable bin sizes where each bin contains an equal number of sources.

While we do not have enough clumps per galaxy to do individual LFs for each galaxy in our sample, we can create a combined function of clumps from all GOALS galaxies. Since we are comparing data from a combined Pα and Pβ sample, we create an SFR function using the SFRs of the clumps as defined previously in Section 4.2. Only extranuclear clumps are included in the SFR function.

We fit the data in two ways to determine the slope. First, we determine the slope by Markov Chain Monte Carlo (MCMC) modeling of the SFR function as a power law plus an exponential cutoff. This simultaneously fits the slope and the incompleteness cutoff point, giving a slope of α = −1.65 ± 0.06 and an SFR cutoff of 0.015 ± 0.002 M_⊙ yr⁻¹ as shown in the left panel of Figure 8. The flattening of the number of sources at lower SFRs is due to an incompleteness in the data at the faint end.

Zoom In Zoom Out Reset image size

Second, we fit a single power law to only the bright end of the SFR function, where the data are complete. We also bin the data using an equal number per bin and determine the error on each bin by bootstrap resampling so we can compare directly to the results from Cook et al. (2016). The SFR cutoff for the fit is determined by the peak of the SFR histogram, at log(SFR cutoff) = –1.7. We find that the combined SFR function of clumps from all GOALS galaxies has a bright-end slope of α = 1.49 ± 0.06.

A study of UV star-forming regions in nearby galaxies by Cook et al. (2016) investigated the dependence of the clump LF on the galaxy SFRs (10⁻³ < SFR [M_⊙ yr⁻¹] < 3). They calculated the SFRs of clumps in their sample by converting dusted-corrected FUV fluxes to SFRs using the Murphy et al. (2011) prescription and Kroupa IMF. Cook et al. (2016) found that galaxies with higher SFRs tend to have flatter clump LF slopes. They then derived a combined LF of clumps from 134 galaxies and found a bright-end slope of α = −1.93. The galaxies in the Cook sample range in distance from ∼1 to ∼10.5 Mpc, corresponding to regions with diameters of ∼24–250 pc. In Figure 8, we also fit only the local normal galaxies having distances larger than 5.8 Mpc (equivalent to a lower limit on the clump resolution of about 70 pc) to match the GOALS sample, and we find a bright-end slope of α = −1.90 ± 0.03, still within the limits of what was initially found with the full sample. This demonstrates that the flattening of the slope that we see in the GOALS sample is not solely due to resolution effects.

The combined SFR function slope from the normal galaxies is steeper than that found in the GOALS galaxies as can be seen in Figure 8. While both studies suffer from incompleteness at the faint end, it is clear that the bright-end slopes are significantly different, suggesting an overabundance of bright, SF extranuclear clumps in LIRGs. Furthermore, the variable nature of extinction in LIRGs and ULIRGs means that some of the clumps in our sample could have A_v as high as 10–20 even if on average the A_v is much lower (Piqueras López et al. 2013, 2016). While we cannot correct for extinction directly in our data, if there was a significant effect, it would imply many more luminous clumps and further flatten the slope of the LF, increasing the difference between the local, normal galaxies and GOALS galaxies.

While the range of sizes of star-forming clumps in local LIRGs is similar to that in local, normal (L_IR < 10¹¹) galaxies, the LIRG clumps have much larger SFR (see Figure 9). Many are comparable to star-forming clumps seen in high-redshift galaxies. A significant fraction of high-z galaxies (1 < z < 3) have turbulent, clumpy, disks with clump masses of ∼10⁸–10⁹ M_⊙ and sizes of 0.5–5 kpc (e.g., Elmegreen et al. 2004, 2009; Daddi et al. 2010). Studying UV star-forming clumps in galaxies in the Hubble Ultra Deep Field at redshifts in the range of 0.5 < z < 1.5, Soto et al. (2017) found an average clump SFR of 0.014 M_⊙ yr⁻¹ and an average clump radius of 0.9 kpc, with some extending up to 2 kpc. The current interpretation is that these large clumps with high SFR have higher gas fractions and increased star formation efficiency (Tacconi et al. 2008; Jones et al. 2010).

Zoom In Zoom Out Reset image size

However, most high-redshift clump studies can only reach down to resolutions of ∼0.5–1 kpc, which is larger than most of the resolved clumps in the GOALS sample. Lensed galaxies provide a way to study high-redshift, clumpy star formation on scales of  100 pc (Livermore et al. 2012, 2015). We can therefore directly compare our GOALS star-forming clumps to those found in the lensed galaxy samples.

The SFRs of GOALS clumps (10⁻³ to 10 [M_⊙ yr⁻¹]) span the range from SINGS z = 0 galaxies to those found in z = 1–3 lensed galaxies, thus bridging the gap between the local universe and the high-z universe when comparing samples at similar resolutions. While local LIRGs have smaller fractions of high-luminosity clumps than galaxies at z > 2, they provide compelling evidence that the physical conditions driving extreme star formation at high z can be found in merging starburst galaxies at low redshift. We note that clumps measured in high-redshift galaxies without the benefit of lensing would lie in the upper right corner of Figure 9 with sizes of 1–5 kpc and SFRs in the range of ∼1–30 M_⊙ yr⁻¹ (Genzel et al. 2011; Swinbank et al. 2012; Wisnioski et al. 2012; Soto et al. 2017; Guo et al. 2018).

Simulations of high-z systems have recently been used to predict the distribution in sizes, luminosities, and lifetimes of the gas clumps in terms of the evolution of gas-rich disks at z ∼ 2 (Oklopčić et al. 2016). It is natural, therefore, to compare our GOALS results to these models to see if they are similar and, if so, whether or not we might gain a better understanding of the nature and fate of the luminous star-forming clumps we see in GOALS.

The MassiveFIRE (Feedback in Realistic Environments) simulation has 8 times smaller mass resolution (m = 3.3 × 10⁻⁴ M_⊙) than FIRE and simulates galaxies at z > 1.7 (Feldmann et al. 2016). The high mass resolution obtained in MassiveFIRE resolves down to an SFR density of 0.07 M_⊙ yr⁻¹ kpc⁻² and allows for a direct comparison to the observed star-forming clumps in GOALS. We have used the same software suite (astrodendro) to identify star-forming clumps in the MassiveFIRE simulation. Using astrodendro on the MassiveFIRE simulations resolves SF clumps down to a size of 126 pc effective radius, allowing for a relevant comparison to our GOALS SF clumps. Figure 10 shows the SFR versus size for both the GOALS and MassiveFIRE star-forming clumps. The clumps in MassiveFIRE have a similar range of SFRs (4 × 10⁻³ to 6.45 M_⊙ yr⁻¹) and sizes (126 pc–1 kpc) to the GOALS star-forming clumps. Since the FIRE simulations model both the stars and gas, we can use these results to estimate basic properties (e.g., the star formation efficiencies) of the star-forming clumps in GOALS.

Zoom In Zoom Out Reset image size

The models predict star-forming clumps with a wide range of total (atomic plus molecular) gas fractions from ∼10% to ∼90% gas. Higher star formation clumps have lower gas fractions. Most of the GOALS clumps that overlap in SFR–size space with MassiveFIRE clumps have intermediate to high clump gas fractions (>50%), while clumps with the highest SFRs, and lowest gas fractions, do appear rare in the GOALS data. The MassiveFIRE clumps have a median star formation efficiency of 3.5 × 10⁻⁹ yr⁻¹, almost an order of magnitude higher than the kiloparsec-scale star formation efficiency found in local spiral galaxies of 4.3 × 10⁻¹⁰ yr⁻¹ (Leroy et al. 2008). Similarly high star formation efficiency has been observed for individual 250 pc regions in one galaxy from our sample, IC 4687, which has both ALMA CO (2–1) and HST Paα data (Pereira-Santaella et al. 2016).

A pilot study with ALMA is currently underway to directly measure the resolved molecular gas content on sizes comparable to our star-forming clumps. These ALMA observations of the GOALS sources will allow us to test for possible correlations with gas content and directly measure the star formation efficiencies of these clumps. In our next paper we will also investigate whether galaxy properties, such as the merger stage, are driving the SFRs in our clumps.