The Dual Role of Starbursts and Active Galactic Nuclei in Driving Extreme Molecular Outflows

We observed the CO (1−0) line in IRAS 20100−4156 using the Atacama Large Millimeter Array (ALMA), over 5 epochs in 2014 June with 37 antennas (PI: Spoon). The absolute flux scale was calibrated using Neptune and Uranus, while ${\rm{J}}2056\mbox{--}4714$ was used as a bandpass and phase calibrator. The receivers were tuned to a central frequency of 102.055 GHz, with a bandwidth of 1.875 GHz in each of four spectral windows, corresponding to 3840 frequency channels with a channel width of 0.49 MHz (∼1.4 km s⁻¹)

CASA $v4.2$ was used for the ALMA pipeline calibration, and the resulting measurement set was used for all further analysis. Continuum subtraction was performed using the task uvcontsub, using line-free channels and spectral windows (a total bandwidth of 2902 MHz), using a first-order polynomial baseline for the continuum. This calibrated and continuum-subtracted visibility data set was used for the task uvmodelfit. Prior to using uvmodelfit, the task statwt was used to calculate the weights for the visibilities based on line-free channels only. The reduced measurement set was imaged and cleaned using the CASA task clean, using natural weighing and a pixel size of 03 × 03, after binning over 71 spectral channels (corresponding to ∼100 km s⁻¹). The resulting cleaned image has an rms noise of 0.1 mJy beam⁻¹ over each channel with width 34 MHz (corresponding to ∼100.2 km s⁻¹ at 102.055 GHz) and a resulting synthesized beam size of 15 × 12 (PA = 75°). As the emission is resolved both in velocity and spatially, the clean mask for the emission region was selected after visual inspection for each channel.

We observed CO (3−2) emission in IRAS 20100−4156, redshifted to ν_obs = 306.082 GHz, using the Atacama Pathfinder Experiment (APEX) single-dish submillimeter telescope over 8 epochs from 2012 August to October (PI: Farrah), for a total on-source time of 3.7 hr. The weather ranged from good (with atmospheric precipitable water vapor (pwv) ∼ 0.5 mm) to acceptable (pwv ∼ 4 mm) during the observations. The APEX-2 heterodyne receiver was used as a front-end, with the eXtended bandwidth Fast Fourier Transform Spectrometer (XFFTS) as the back-end, with wobbler switching at a frequency of 0.5 Hz and an asymmetrical azimuthal throw of 50''. Saturn and RZ-Sgr were used for pointing. The total bandwidth was 2.5 GHz, centered at 306.0119 GHz, with a frequency resolution of 76.3 kHz, which corresponds to a velocity resolution of 0.074 km s⁻¹. The half-power beam width at this frequency is ∼21'', so the emission was unresolved. Data reduction was carried out using GILDAS class (Continuum and Line Analysis Single-dish Software), and the spectrum was binned to a velocity resolution of ∼53 km s⁻¹. The antenna temperatures (${T}_{{\rm{A}}}$) were converted to flux densities assuming an aperture efficiency of ${\eta }_{{\rm{a}}}=0.6$, which implies a K-to-Jy conversion factor of $24.4/{\eta }_{{\rm{a}}}=41$ Jy K⁻¹.¹¹ The final spectrum has an rms noise of 0.6 mJy over a line-free bandwidth of ∼3300 km s⁻¹.

We observed CO (1−0) emission toward IRAS 03158+4227 with the Plateau de Bure Interferometer (PdBI) in the most compact five-antenna configuration (5D), during 2013 July–August (PI: Spoon). The weather was average during these observations (pwv ∼ 10 mm on average). The quasars 3C 84 and B0234+285 were used as phase and amplitude calibrators, respectively, and the quasars 3C 84 and 3C 345 were used as RF bandpass calibrators. The star MWC 349 and the quasar 3C 454.3 were used as absolute flux calibrators. The WideX correlator with a bandwidth of 3.6 GHz was tuned to 101.604 GHz, and the observations were carried out in dual-polarization mode, with a spectral resolution of 1.95 MHz (corresponding to ∼5.8 km s⁻¹). The total on-source time, after flagging of bad visibilities, was 19.5 hr (six-antenna equivalent time, summing together 18 observations). Given the large bandwidth of the observations, we also covered CN (N = 1–0, J = 3/2–1/2), redshifted from ${\nu }_{\mathrm{rest}}\sim 113.490\,\mathrm{GHz}$ to ${\nu }_{\mathrm{obs}}\,=100.04612\,\mathrm{GHz}$.

All observations were calibrated using the IRAM PdBI data reduction pipeline in clic (Continuum and Line Interferometer Calibration), with subsequent additional flagging by hand. The absolute flux scale was calibrated to better than 20%. Continuum subtraction was performed for the reduced visibility cube using the task uv_subtract, excluding the line channels (a total of 2.64 GHz of line-free bandwidth). The resulting continuum-subtracted visibility data were imaged using uv_map, with natural weighing and a pixel size of 05 × 05, and cleaned using the task clean with the Hogbom algorithm at a binned spectral resolution of 20 MHz (∼60 km s⁻¹ at a frequency of 101.604 GHz). The resulting cleaned image has an rms noise of 0.26 mJy beam⁻¹ over each 20 MHz channel and a beam size of 48 × 39 (PA = 166°). Primary-beam correction was applied using the task primary.

IRAS 20100−4156 is a two-component interacting system, and we here refer to the two principal components as A and B, respectively (Figure 1). The two components have a projected separation of ∼25, corresponding to 5.6 kpc (1'' ∼ 2.287 kpc at z = 0.12975). We detect and spatially resolve the CO (1−0) line emission between both components in the continuum-subtracted spectral line cube. The underlying continuum flux is ${f}_{\lambda }=1.56\pm 0.03$ mJy at ${\lambda }_{\mathrm{obs}}=2.96\,\mathrm{mm}$, using the line-free channels (Figure 1). The dust emission predominantly emerges from component A, though we tentatively detect continuum emission at ∼3σ significance from component B, with a flux of ${f}_{\lambda }=93\pm 36$ μJy. We fit the line centroids in spectra extracted from the peak pixels of A and B with 1D Gaussian line profiles (Figure 2) and find velocity widths of ${\rm{\Delta }}{v}_{\mathrm{FWHM}}^{{\rm{A}}}\sim 371.4\pm 4.2$ km s⁻¹ and ${\rm{\Delta }}{v}_{\mathrm{FWHM}}^{{\rm{B}}}\sim 111.8\pm 8.2$ km s⁻¹, respectively, with both lines centered at v ∼ 0 km s⁻¹. We find no significant velocity/redshift offset between the two components, indicating that they do not show any motion relative to each other. This is also seen in the channel maps for the line core emission (Figure 3) and in the moment-1 map (Figure 4, panel (b)), which does not show a significant velocity gradient between A and B components.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The CO moment-0 map (Figure 1) is created by integrating over the FWZI of the CO line. We fit the A and B component emission in the CO (1−0) moment-0 maps with 2D Gaussian components and find velocity-integrated line fluxes of ${I}_{\mathrm{CO}}^{{\rm{A}}}\,=10.7\pm 0.4$ Jy km s⁻¹ and ${I}_{\mathrm{CO}}^{{\rm{B}}}=1.8\pm 0.1$ Jy km s⁻¹ for components A and B, respectively. These correspond to CO line luminosities of ${L}_{\mathrm{CO},{\rm{A}}}^{{\prime} }=(8.4\pm 0.3)\times {10}^{9}$ K km s⁻¹ pc⁻² and ${L}_{\mathrm{CO},{\rm{B}}}^{{\prime} }=(1.4\pm 0.1)\times {10}^{9}$ K km s⁻¹ pc⁻², respectively. Assuming a CO-to-H₂ gas mass conversion factor ${\alpha }_{\mathrm{CO}}\,=0.8$ M_⊙ (K km s⁻¹ pc⁻²)⁻¹, suitable for ULIRGs in the local universe (see Bolatto et al. 2013, for a review), we find total H₂ gas masses of ${M}_{{\rm{A}}}=(6.7\pm 0.2)\times {10}^{9}{M}_{\odot }$ and ${M}_{{\rm{B}}}\,=(1.1\pm 0.1)\times {10}^{9}{M}_{\odot }$ for components A and B, respectively.

We detect outflowing molecular gas from component A in the form of significant emission beyond $v\sim \pm 450$ km s⁻¹ (Figure 2), also seen in the position–velocity (PV) diagram (Figure 4). We discuss the outflow wing emission in greater detail in the next section.

The determination of the outflow properties depends sensitively on the velocity widths (ranges) assumed for the high-velocity outflow wings, and we define the velocity ranges for outflow wings in an iterative manner. The initial spectrum is extracted from a 5'' × 5'' region centered on the peak pixel, and we simultaneously fit three Gaussian profiles to the spectrum, one each for the red wing, the blue wing, and the line core, to avoid assumptions about the symmetry of the red and blue wings. The inner limits of the outflow velocity width are defined as the channel at which the wing emission starts dominating over the core emission, and the outer limits are defined using the fitted broad Gaussian components. The cleaned image cube is binned over these velocity ranges to obtain emission maps for the red and blue wings, and CO spectra are re-extracted from the peak emission pixel in the red and blue emission maps. We again fit the obtained spectrum with a sum of three Gaussian components and repeat the same procedure as above until it converges. We bin the calibrated and continuum-subtracted uv-visibility data over the final velocity ranges and use model fitting to the binned visibilities for the red and blue wings to determine the outflow positions and fluxes. We fit models to the binned visibility data using the CASA task uvmodelfit, assuming a 2D Gaussian source to obtain the FWHM spatial extent and the flux (see the Appendix for details).

We find the velocity ranges for the red and blue outflow wings to be $\in (440,1650)$ km s⁻¹ and $\in (-1169,-466)$ km s⁻¹, respectively. Based on model fitting of the visibility data binned over these velocity ranges, we find integrated fluxes of ${I}_{\mathrm{CO}}^{\mathrm{red}}=0.54\,\pm 0.03$ Jy km s⁻¹ and ${I}_{\mathrm{CO}}^{\mathrm{blue}}=0.58\pm 0.03$ Jy km s⁻¹ and measure a velocity-integrated line flux of ${I}_{\mathrm{CO}}^{\mathrm{core}}=11.2\pm 0.7$ Jy km s⁻¹ in the line core. We find a projected spatial offset of 09 ± 01 between the peaks of the red and blue emission, which corresponds to a physical distance of ∼2.0 ± 0.2 kpc, which we define as the outflow diameter (Table 3). The peaks of the red and blue wings are spatially offset from the peak CO (1−0) and dust emission by 03 ± 005 and 06 ± 01, corresponding to physical distances of 0.7 ± 0.1 kpc and 1.4 ± 0.2 kpc, respectively (Figure 4). The resultant properties are consistent with those calculated from the cleaned image. The final velocity ranges and fluxes are given in Table 3. The outflow appears to be aligned with the velocity gradient seen across the galaxy (Figure 4), although the velocity gradient is small between the peaks of the red and blue outflow wing. This raises the intriguing possibility that rotational support may play a role in the origin of the most powerful outflows (assuming that the outflow is in the plane of the galaxy; see González-Alfonso et al. 2017).

We detect CO (3−2) line emission at a ∼21σ significance from IRAS 20100−4156 (Figure 2). The A and B components are spatially unresolved in the observations. Using a Gaussian fit to the observed spectrum, we find a line width of ${\rm{\Delta }}{v}_{\mathrm{FWHM}}=293\pm 25$ km s⁻¹. Integrating the line emission over the FWZI (v ∼ −269 km s⁻¹ to $v\sim 317$ km s⁻¹), we find an integrated flux of ${I}_{\mathrm{CO}(3\mbox{--}2)}=85.8\pm 4.0$ Jy km s⁻¹. This corresponds to a line luminosity of ${L}_{\mathrm{CO}\ (3\mbox{--}2)}^{{\prime} }=(7.5\pm 0.4)\,\times {10}^{9}$ K km s⁻¹ pc⁻². Using the total observed CO (1−0) line luminosity (for both the A and B components combined), we find a brightness temperature ratio of ${r}_{31}\sim 0.8$, consistent with the mean ratio in other local ULIRGs ($\langle {r}_{31}\rangle \sim 0.7$; Papadopoulos et al. 2012).

We do not detect the high-velocity outflow wings seen in CO (1−0) in CO (3−2) emission at the depth of these observations. However, we tentatively find that while normalized line profiles for CO (3−2) and CO (1−0) agree at $v\gt 0$ km s⁻¹, they show a mismatch at $v\lt 0$ km s⁻¹ (Figure 2(d)). This can be quantified using the line luminosity ratio ${L}_{\mathrm{CO}\ (3\mbox{--}2)}^{{\prime} }/{L}_{\mathrm{CO}(1\mbox{--}0)}^{{\prime} }={r}_{31}$ for the red- and blueshifted emission in the line core. We find ${r}_{31}^{\mathrm{blue}}=0.6\pm 0.2$, while ${r}_{31}^{\mathrm{red}}=0.9\pm 0.2$, where we have selected only those channels where CO (3−2) emission is detected at $\gtrsim 2\sigma$ significance. While both ${r}_{31}^{\mathrm{red}}$ and ${r}_{31}^{\mathrm{blue}}$ are consistent with the median r₃₁ in ULIRGs, there is no physical reason for a difference in excitation between the red- and blueshifted emission. The observations can be explained if CO (3−2) emission is under-luminous on the blueshifted side as compared to the redshifted emission. As the CO (3−2) emission traces more compact nuclear emission than CO (1−0), self-absorption in the molecular gas can result in decreased flux, leading to a lower ${r}_{31}^{\mathrm{blue}}$ than ${r}_{31}^{\mathrm{red}}$. We therefore conclude that the average ${r}_{31}\gtrsim 0.8$ and treat it as a lower limit on the gas excitation.

We detect strong OHM emission from IRAS 20100−4156 at a ∼40σ significance. We detect line emission from a velocity range of −927 to 527 km s⁻¹, with tentative evidence of absorption redward of 527 km s⁻¹ (Figure 2). The OHM emission could be contaminated by the OH 1.665 GHz satellite line (OH ${}^{2}{{\rm{\Pi }}}_{3/2}$, $J=3/2,F={1}^{e}-{1}^{f}$) at the expected position of $v\sim +360$ km s⁻¹, but we do not detect excess emission at that velocity. As the emission is spatially unresolved, we extract the spectrum from the peak pixel of the moment-0 map (Figure 2). Based on a 2D elliptical Gaussian fit to the moment-0 map, we find an integrated line flux of ${I}_{\mathrm{OH}}=27.9\pm 0.8$ Jy km s⁻¹. This corresponds to a line luminosity of ${L}_{\mathrm{OH}}=(1.5\pm 0.1)\times {10}^{4}{L}_{\odot }$. We find a systematic velocity offset of $\delta v\sim 190\pm 7$ km s⁻¹ between the CO (1−0) and OHM lines, as well as a spatial offset of 03 ± 01 between the peaks of emission. We discuss these differences further in Section 5.3.

IRAS 03158+4227 is a merging system with two distinct galaxies spatially offset by 189, corresponding to 45.4 kpc (1'' ∼ 2.360 kpc at z = 0.13459), as well as tidal structure with two components, one connecting the two galaxies. The two principal components are labeled A and B, while the two tidal components are labeled T and E (Figure 5). We detect and spatially resolve the CO (1−0) line emission between each of these, and we detect high-velocity outflowing molecular gas from component A (Figure 6). We do not find a significant velocity offset between any of the components, as can be seen in the channel maps of the CO emission (Figure 7), and the moment-1 map of the emission shows a relatively smooth velocity gradient across all components (Figure 8). The moment-0 map of the CO emission is made by integrating over the FWZI line width (Figure 5). The continuum emission, restricted to component A, has a flux of ${f}_{\lambda }=0.94\pm 0.03$ mJy at ${\lambda }_{\mathrm{obs}}=3\,\mathrm{mm}$, using the line-free channels.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The velocity widths and spatial extents of components A, B, T, and E are significantly different. We therefore extract the fluxes for each component in the following iterative manner: we select a 3'' × 3'' region around the A, B, and E components (and a 9'' × 3'' region around the elongated tidal tail T). We extract spectra from each of these regions, which is well described by a Gaussian line profile. Based on these spectra, we integrate over all contiguously positive channels around the line center to obtain a moment-0 map for each region. We then fit each component with a 2D elliptical Gaussian and thus extract the flux from each component—A, B, T, and E. We iteratively fit and subtract emission from A, B, and T and find velocity-integrated line fluxes of ${I}_{\mathrm{CO}}^{{\rm{A}}}=13.1\pm 0.1$, ${I}_{\mathrm{CO}}^{{\rm{B}}}=2.8\pm 0.3$, ${I}_{\mathrm{CO}}^{{\rm{T}}}=2.1\pm 0.4$, and ${I}_{\mathrm{CO}}^{{\rm{E}}}=1.2\pm 0.1$ Jy km s⁻¹, respectively, for each of the components. These imply line luminosities of ${L}_{\mathrm{CO},{\rm{A}}}^{{\prime} }=(1.1\pm 0.1)\,\times {10}^{10}$, ${L}_{\mathrm{CO},{\rm{B}}}^{{\prime} }\,=\,(2.4\,\pm \,0.2)\times {10}^{9}$, ${L}_{\mathrm{CO},{\rm{T}}}^{{\prime} }=(1.8\pm 0.4)\times {10}^{9}$, and ${L}_{\mathrm{CO},{\rm{E}}}^{{\prime} }=(9.7\pm 0.8)\times {10}^{8}$ K km s⁻¹ pc⁻², respectively. Assuming a CO-to-H₂ conversion factor of ${\alpha }_{\mathrm{CO}}=0.8$ M_⊙ (K km s⁻¹ pc⁻²)⁻¹ for A and B and ${\alpha }_{\mathrm{CO}}=3.6$ M_⊙ (K km s⁻¹ pc⁻²)⁻¹ for the tidal components, we find gas masses of ${M}_{{\rm{A}}}=(8.9\pm 0.8)\times {10}^{9}$, ${M}_{{\rm{B}}}=(1.9\pm 0.2)\times {10}^{9}$, ${M}_{{\rm{T}}}\,=(6.3\pm 1.3)\times {10}^{9}$, and ${M}_{{\rm{E}}}=(3.6\pm 0.5)\times {10}^{9}{M}_{\odot }$, respectively. The differing conversion factors are physically motivated, with the lower conversion factor suitable for ULIRGs, which typically show higher gas temperatures, while the higher α_CO is more realistic for molecular gas in Milky-Way-type giant molecular clouds and therefore is more suitable for the tidal tails. We find a gas mass ratio of ∼5:1 between the two principal components A and B. We detect the high-velocity wing emission at velocities beyond ±1000 km s⁻¹ in the PV diagram for IRAS 03158+4227 (Figure 8).

We obtain the velocity widths for the outflow wings in the same manner as for IRAS 20100−4156 (see Section 3.1.2) and find the velocity range $\in (400,1750)$ km s⁻¹ and $\in (-944,-413)$ km s⁻¹ for the red and blue outflow wings, respectively, for IRAS 03158+4227. Based on uv model fitting of the binned visibility data over this velocity range, we find integrated fluxes of ${I}_{\mathrm{CO}}^{\mathrm{red}}=0.80\pm 0.22$ Jy km s⁻¹ and ${I}_{\mathrm{CO}}^{\mathrm{blue}}=0.31\pm 0.15$ Jy km s⁻¹ in the red and blue wings, respectively. We measure a total velocity-integrated line flux of ${I}_{\mathrm{CO}}^{\mathrm{core}}=8.4\pm 0.7$ Jy km s⁻¹ in the line core. We find a spatial offset of 16 ± 11 between the peaks of the red and blue emission, which corresponds to a physical distance of 3.9 ± 2.7 kpc (which we use as the outflow diameter). The peaks of the red and blue wings are spatially offset from the CO (1−0) and dust peaks by 06 ± 02 and 11 ± 05, respectively, corresponding to a physical distance of ∼1.4 ± 0.4 kpc and ∼2.6 ± 1.2 kpc (Figure 8). The difference in the red and blue outflow peak offsets indicates a significant asymmetry in the outflow morphology. The outflow is counter-aligned with the galaxy-wide velocity gradient (Figure 8), in direct contrast with that seen in IRAS 20100−4156. It is therefore unlikely that host galaxy rotation is an essential component in launching the outflows.

CN abundance is enhanced in UV-irradiated surfaces of molecular gas clouds and therefore preferentially traces photodissociated regions (PDRs; Rodriguez-Franco et al. 1998), while CO traces the entirety of the molecular gas (dense as well as diffuse components). A low ${L}_{\mathrm{CO}}^{{\prime} }/{L}_{\mathrm{CN}}^{{\prime} }$ ratio therefore indicates relatively more PDR/XDR-dominated molecular gas.

We detect CN (N = 1–0) line emission from only component A in IRAS 03158+4227 Figure 5. While the line is a doublet, with the fine-structure components (J = 1/2–1/2) and (J = 3/2–1/2) separated by $\delta {\nu }_{\mathrm{rest}}\sim 0.31\,\mathrm{GHz}$, corresponding to $\delta v\sim 819$ km s⁻¹, we cover only the (J = 3/2–1/2) component of the transition. Fitting a 1D Gaussian to the spectral line, we find a velocity width of ${\rm{\Delta }}{v}_{\mathrm{FWHM}}=312.5\pm 24.9$ km s⁻¹, centered on ${v}_{\mathrm{sys}}=-42.9\pm 10.4$ km s⁻¹. We create a moment-0 map by integrating over the FWZI velocity range (v = −289 km s⁻¹ to v = 429 km s⁻¹). Fitting the moment-0 emission with a 2D Gaussian source, we find an integrated line flux of ${I}_{\mathrm{CN}}=1.0\pm 0.1$ Jy km s⁻¹. This corresponds to a line luminosity of ${L}_{\mathrm{CN}}^{{\prime} }=(8.9\pm 0.5)\times {10}^{8}$ K km s⁻¹ pc⁻².

Under conditions of local thermal equilibrium (LTE) between the two fine-structure components, we expect that the two components will have a ratio of

$$\begin{eqnarray}&&{R}_{12}\mathrm{CN}(N=1\mbox{--}0)=\displaystyle \frac{I(J=1/2\mbox{--}1/2)}{I(J=3/2\mbox{--}1/2)}=0.5.\end{eqnarray} \tag{ 1 }$$

The expected line profile of the doublet in LTE is shown in Figure 6. While the CN (N = 1–0, J = 3/2–1/2) and (N = 1–0, J = 1/2–1/2) lines are split further into hyperfine-structure lines, we do not spectrally resolve them. Assuming LTE, we find an expected integrated line flux of ${I}_{\mathrm{CN}}=1.5\,\pm 0.2$ Jy km s⁻¹ for both components together, corresponding to a line luminosity of ${L}_{\mathrm{CN}}^{{\prime} }=(1.3\pm 0.1)\times {10}^{9}$ K km s⁻¹ pc⁻². Using the CO (1−0) line luminosity as calculated above, we find a line luminosity ratio of ${L}_{\mathrm{CO}}^{{\prime} }/{L}_{\mathrm{CN}}^{{\prime} }\sim 8.2\pm 0.5$. The observed ${L}_{\mathrm{CO}}^{{\prime} }/{L}_{\mathrm{CN}}^{{\prime} }$ is consistent with the globally averaged line ratios seen in other ULIRGs and is similar to that seen in Mrk 231 (${L}_{\mathrm{CO}}^{{\prime} }/{L}_{\mathrm{CN}}^{{\prime} }\sim 8$; Aalto et al. 2002; Pérez-Beaupuits et al. 2007). With a critical density 5 times lower than HCN (1−0), CN (N = 1–0) does not trace the same molecular gas as HCN, and ${L}_{\mathrm{HCN}}^{{\prime} }/{L}_{\mathrm{CN}}^{{\prime} }$ varies in the range of ∼0.5–6 for galaxies with a constant ${L}_{\mathrm{CO}}^{{\prime} }/{L}_{\mathrm{HCN}}^{{\prime} }$ (Aalto et al. 2002). We therefore do not use it to calculate the dense gas mass.

We detect spatially unresolved OHM emission from component A of IRAS 03158+4227 (Figure 6). Fitting a 1D Gaussian to the spectral line profile, we find a velocity width of ${\rm{\Delta }}{v}_{\mathrm{FWHM}}\sim 803\pm 110$ km s⁻¹ and a velocity offset of ${v}_{0}\,\sim -252\pm 46$ km s⁻¹ between the CO and OHM lines. Based on a 2D elliptical Gaussian fit to the moment-0 map (Figure 5), we find an integrated line flux of ${I}_{\mathrm{OH}}=4.1\pm 0.8$ Jy km s⁻¹. This corresponds to a line luminosity of ${L}_{\mathrm{OH}}=(2.4\pm 0.5)\,\times {10}^{3}{L}_{\odot }$. We detect a marked asymmetry in the maser emission line profile, with a significant excess in the blueshifted emission, similar to that observed in the OHM spectrum for IRAS 20100−4156. This is discussed further in Section 5.3.

We perform SED fitting for both sources using CIGALE (Code Investigating GAlaxy Emission; Noll et al. 2009; Serra et al. 2011). CIGALE builds galaxy SEDs from UV to radio wavelengths assuming a combination of modules and summing the emission from each, and it allows modeling the star formation history (SFH), the stellar emission using population synthesis models (Bruzual & Charlot 2003; Maraston 2005), nebular lines, dust attenuation (e.g., Calzetti et al. 2000), dust emission (e.g., Draine & Li 2007; Casey 2012), contribution from AGNs (Fritz et al. 2006; Dale et al. 2014), and radio synchrotron emission. The SEDs implicitly maintain the energy budget, with consistency between UV dust attenuation and the dust emission. For our SED fitting, we assume the SFH to be a double exponentially decreasing function of time, the dust attenuation as in Calzetti et al. (2000), and the dust emission models from Draine et al. (2014).

We use all available photometric data points (Table 2) for each of the sources¹³ as described in Section 2.3. We add multiple line-free continuum points from the IRS spectra/PACS spectroscopy to our SED fitting, to ensure consistency between the spectroscopic and the SED fit observations.

The photometry resolves the targets and their merging/interacting companions in some cases and not in others. For IRAS 20100−4156, the A and B components are not dynamically independent and are only spatially resolved in the HST images. We therefore treat them as one system. We re-extract HST photometry so as to include the contribution from both A and B components, and then we perform the SED fitting. Based on the modest detection of the 3 mm dust continuum from component B, we expect that the bulk of the far-IR emission is from component A. For IRAS 03158+4227, the A and B components are separated by a projected distance of ∼22'' and spatially resolved in all images except Herschel SPIRE photometry. We therefore perform the SED fitting for A with and without the SPIRE photometry, to test for contamination from component B, and find no significant differences in our results. The near-IR wavelengths for IRAS 03158+4227 (<6 μm) also show contamination from its neighboring star, and we re-extract the photometry where possible. Finally, CIGALE performs a probability distribution function analysis for our specified module parameters and obtains the likelihood-weighted mean value for each. The best-fit SED models are shown in Figure 9.

Zoom In Zoom Out Reset image size

We test for any AGN contribution to the SEDs using the Fritz et al. (2006) module, which includes AGN emission reprocessed by a dusty torus, with the torus opening angle, orientation, and inner and outer radii being free parameters (see Fritz et al. 2006, for further details). We extensively test for AGN contributions to the IR luminosity by varying the initial conditions over the entire allowed parameter space for dusty torus models. We find bolometric AGN luminosities of ${L}_{\mathrm{AGN}}=(7.2\pm 0.3)\times {10}^{11}{L}_{\odot }$ and ${L}_{\mathrm{AGN}}=(1.3\pm 0.1)\times {10}^{12}{L}_{\odot }$ for IRAS 20100−4156 and IRAS 03158+4227, respectively, and ${L}_{\mathrm{IR}}=3.7\times {10}^{12}{L}_{\odot }$ and ${L}_{\mathrm{IR}}=3.8\times {10}^{12}{L}_{\odot }$. Together these correspond to AGN fractions of f_AGN = 0.2 ± 0.1 and f_AGN = 0.3 ± 0.1 for IRAS 20100−4156 and IRAS 03158+4227, respectively, indicating that neither of the sources is AGN-dominated. The final parameters estimated from the SED fitting—SFR, dust mass, and stellar mass—are listed in Table 5.

We find a gas-to-dust mass ratio ${\delta }_{\mathrm{GDR}}\sim 10-20$ for both IRAS 20100−4156 and IRAS 03158+4227. Previous values for ${\delta }_{\mathrm{GDR}}$ in (U)LIRGs calculated using galaxy-wide SED fitting to obtain the dust mass are similarly low (${\delta }_{\mathrm{GDR}}\sim 8\mbox{--}20;$ Papadopoulos et al. 2010). These values are significantly smaller than those observed in nearby normal, star-forming galaxies (${\delta }_{\mathrm{GDR}}\sim 100\mbox{--}200$; Rémy-Ruyer et al. 2014). In addition, such low ${\delta }_{\mathrm{GDR}}$ values are hardly consistent with the strong far-IR molecular absorption observed in the far-IR (e.g., González-Alfonso et al. 2015), which favor Milky-Way-like values with molecular abundances consistent with chemical models (González-Alfonso et al. 2018). Furthermore, spatially resolved observations of nuclear regions in (U)LIRGs also show Milky-Way-like high ${\delta }_{\mathrm{GDR}}$ (e.g., Wilson et al. 2008), instead of the extremely low values inferred in ULIRGs.

This apparent contradiction is due to the degeneracy between the dust optical depth and the dust temperature. Emission from dust that is optically thick up to far-IR or submillimeter wavelengths (e.g., Sakamoto et al. 2008; González-Alfonso et al. 2015) naturally leads to excess far-IR/submillimeter emission and can dominate the emission from the cold dust component, leading to an overestimate of the dust mass. Distinguishing between these two cases is only possible with spatially resolved observations of the dust continuum in the long-wavelength, optically thin regime.

Mid-IR spectroscopy offers a powerful way to determine the AGN fraction even in dust-obscured nuclei, as the blackbody emission from AGN-heated dust at T ∼ 1000–1500 K at the inner boundary of the dust torus (the dust sublimation temperature) within ∼10 pc of the AGN peaks in the mid-IR. For IRAS 20100–4516 and IRAS 03158+4227, we do not detect broad-line signatures in optical/near-IR/mid-IR spectra, suggesting that we are looking at the torus edge-on. We calculate the AGN fraction using the method developed in Nardini et al. (2008), based on the observation that the relative difference between mid-IR spectra of AGNs and starburst-dominated galaxies is the highest in the 5–8 μm regime. The 5–8 μm mid-IR spectra of starburst galaxies are fairly consistent (Brandl et al. 2006) and are dominated by emission from polycyclic aromatic hydrocarbon (PAH) features at 6.2 and 7.7 μm, while those of AGN-dominated galaxies show a smooth power-law continuum emission. PAH features are associated with star formation activity, as they exist in photodissociation regions and H ii regions, and are destroyed in the intense radiation fields surrounding an AGN (e.g., Genzel et al. 1998; Laurent et al. 2000). The mid-IR spectra of ULIRGs can therefore be decomposed into their AGN and starburst components in the following manner:

$$\begin{eqnarray}&&{f}_{\nu }^{\mathrm{obs}}={f}_{\nu }^{\mathrm{int}}[(1-{\alpha }_{6}){u}_{\nu }^{\mathrm{SB}}+{\alpha }_{6}{u}_{\nu }^{\mathrm{AGN}}{e}^{-\tau (\lambda )}],\end{eqnarray} \tag{ 8 }$$

where ${f}_{\nu }^{\mathrm{obs}}$ is the observed flux, ${f}_{\nu }^{\mathrm{int}}$ is the intrinsic flux density, ${u}_{\nu }^{\mathrm{SB}}$ and ${u}_{\nu }^{\mathrm{AGN}}$ are the SB and AGN templates, normalized at 6 μm, ${\alpha }_{6}$ is the fraction of the flux contributed by the AGN at 6 μm, and the PAH features are automatically included in the SB template. We include the dust extinction in the form of $\tau (\lambda )\propto {\lambda }^{-1.75}$ (Draine & Li 2007). There are additional features detected in the mid-IR spectra of ULIRGs, such as the 6 μm water-ice absorption, and additional aliphatic hydrocarbon (HAC) features at 6.85 and 7.25 μm. We include the 6 μm water-ice feature using the 15 K pure water-ice spectrum from the Leiden ice database,¹⁴ and we model the 6.85 and 7.25 μm aliphatic hydrocarbon features as Gaussians. We apply these to the fitting by adding additional multiplicative terms to Equation (8). We create the starburst template by averaging the mid-IR spectra from five starburst-dominated ULIRGs, IRAS 10190+1322, IRAS 12112+0305, IRAS 17208−0014, IRAS 20414−1651, and IRAS 22491−1808 (all with f_AGN ≲ 0.05, using multi-band observations in the literature), and use a power-law template for the AGN contribution (${f}_{\nu }\propto {\lambda }^{\gamma }$, with γ ∼ 0.7–1.5). We leave the parameters for the ice features, ${\alpha }_{6}$, the optical depth at 6 μm, and the overall normalization as free parameters in the fitting. We extrapolate from the ${\alpha }_{6}$ to f_AGN, the AGN contribution to the bolometric luminosity, i.e., ${L}_{\mathrm{AGN}}\sim {f}_{\mathrm{AGN}}{L}_{\mathrm{IR}}$, and thus the AGN luminosity L_AGN (see Nardini et al. 2008, 2010, for more details).

The best-fit mid-IR decompositions for IRAS 20100−4156 and IRAS 03158+4227 are shown in Figure 10. We find that the best-fit model deviates from the observed mid-IR emission at 5–5.5 μm and produces a shallower slope in that spectral range than is actually observed. The steeper slope observed is likely due to the extended 4–5 μm CO gas/ice absorption feature (see Spoon et al. 2004), which we have not included in our modeling. We find AGN fractions of f_AGN = 0.14 ± 0.47 and f_AGN = 0.43 ± 0.18 for IRAS 20100−4156 and IRAS 03158+4227, respectively. These are consistent within the uncertainties with f_AGN found using SED fitting (Section 4.2.1) and also consistent with the AGN fractions found previously for these objects (Nardini et al. 2009) and those typically found in ULIRGs (Farrah et al. 2012).

Zoom In Zoom Out Reset image size

To enable comparison between IRAS 20100−4156, IRAS 03158+4227, and other molecular outflows detected using CO in ULIRGs, we have compiled all such sources from the literature (Cicone et al. 2014; García-Burillo et al. 2015; Veilleux et al. 2017) and also performed a similar mid-IR decomposition for them. We find results generally consistent with previous AGN fractions for those sources included in the Cicone et al. (2014) sample.¹⁵ The results of all these are shown in Table 4.

We compare L_AGN from mid-IR spectral decomposition to the previously obtained L_AGN from X-ray observations for our two sources. Primary hard X-ray spectra from AGNs are characterized by an X-ray photon index Γ (where the number of photons at energy E is given by $n(E)\propto {E}^{-{\rm{\Gamma }}}$), with the power-law cutoff energy determined by the column density ${N}_{{\rm{H}}}$, and increasing to higher photon energies at higher column densities. In addition to this, X-ray spectra from AGN-dominated ULIRGs also show a broad Fe Kα line at 6.4 keV, with equivalent width EW ∼ 1 keV. The soft X-ray emission (≲1 keV) is often dominated by thermal emission from a hot plasma at ∼0.7 keV from nuclear starburst activity. While primary hard X-ray emission from the AGN may be blocked at energies <10 keV in deeply obscured Compton-thick sources (${N}_{{\rm{H}}}\gtrsim 1.5\times {10}^{24}$ cm⁻²), spectra up to 10 keV can be used to determine the power-law cutoff (which gives the column density), the Fe Kα line, and reflection components, depending on the geometry of the absorber.

For IRAS 20100−4156, XMM-Newton was used to observe the 0.5–10 keV X-ray emission (Franceschini et al. 2003). The emission was detected at modest significance, and the spectrum was modeled using a combination of thermal plus a power-law emission (either absorbed or with a cutoff, both similarly good fits). The photon indices were fixed to ${\rm{\Gamma }}\sim 1.1$ and ${\rm{\Gamma }}\sim 1.7$, and both fits gave comparable values for $({N}_{{\rm{H}}}\,\sim (2-2.3)\times {10}^{22})$ cm⁻² and intrinsic absorption-corrected 2–10 keV luminosity of ${L}_{2-10\mathrm{keV}}=1.6\times {10}^{42}$ erg s⁻¹ i.e., IRAS 20100−4156 does not show AGN-dominated X-ray emission. AGN-host galaxies also show a relative far-IR deficit (or X-ray excess) in the X-ray–${L}_{\mathrm{FIR}}$ relation (Franceschini et al. 2003; Ranalli et al. 2003). IRAS 20100−4156 is consistent within the scatter in this relation, showing that the observed X-ray emission is predominantly from starburst activity.

However, we note that the obtained column density is lower than the column density obtained from modeling of far-IR OH lines (González-Alfonso et al. 2017), which is >10²⁴ cm⁻². It is then possible that IRAS 20100−4156 hosts an intrinsically X-ray weak and deeply obscured AGN, which is drowned out by the X-ray emission from less obscured circumnuclear starbursts. Such X-ray-weak AGNs have been seen in local ULIRGs (including Mrk 231; Teng et al. 2014, 2015) and have been attributed to super-Eddington accretion onto the central SMBH (Vasudevan & Fabian 2009; Lusso et al. 2010, 2012).

A similar lack of X-ray AGN signatures is found for IRAS 03158+4227. The Advanced Satellite for Cosmology and Astrophysics (ASCA) was used to observe the hard X-ray emission in IRAS 03158+4227 and detected a 2–10 keV flux of ∼(5.4 ± 1) × 10⁻¹³ erg s⁻¹ cm⁻² (Risaliti et al. 2000). This implies a 2–10 keV luminosity of ${L}_{2-10\mathrm{keV}}=(2.5\pm 0.5)\,\times {10}^{43}$ erg s⁻¹, consistent with thermal emission, though Risaliti et al. (2000) invoke high column densities of ${N}_{{\rm{H}}}\gt {10}^{24}$ cm⁻² to explain the lack of X-ray AGN signatures. We also note that the ASCA field of view is $50^{\prime}$, and the detected emission is likely contaminated by the interacting companion of IRAS 03158+4227, which is a Seyfert 1 galaxy (Meusinger et al. 2001).

Overall we do not find X-ray signatures for AGNs in either IRAS 20100−4156 and IRAS 03158+4227. We convert between ${L}_{2-10\mathrm{keV}}$ and the bolometric AGN luminosity using relations from Marconi et al. (2004) and find ${L}_{\mathrm{AGN}}\,\sim 2\times {10}^{10}{L}_{\odot }$ and ${L}_{\mathrm{AGN}}\sim 3\times {10}^{11}{L}_{\odot }$, which correspond to f_AGN ∼ 0.005 and f_AGN ∼ 0.08 for IRAS 20100−4156 and IRAS 03158+4227, respectively. The ratios of the ${L}_{2-10\mathrm{keV}}$ to the bolometric AGN luminosity are then 0.08% and 0.4%, respectively. These are significantly lower than what is typically seen for Seyfert 1 galaxies (ratios between 2% and 15%; Elvis et al. 1994) and are consistent with both sources being extremely under-luminous in X-ray emission, similar to that seen in other ULIRGs (Teng et al. 2015), which has been attributed to higher Eddington ratios. Future hard X-ray observations of the two sources at energies >10 keV are needed to confirm this hypothesis. Meanwhile, we treat the L_AGN from X-ray observations as lower limits on the AGN luminosity.

We also mention other commonly used optical/near-IR/mid-IR AGN diagnostics for completeness. These include high-ionization mid-IR lines—e.g., [Ne v] 14.3 μm and [Ne vi] 7.6 μm—which are only detected in AGN-host sources, and the optical lines [O iii]/Hβ, whose ratio is used to characterize line ionization and the hardness of the radiation field.

Neither IRAS 20100−4156 nor IRAS 03158+4227 displays the high-ionization lines, which are most commonly seen in AGN-host galaxies, though they display the lower-energy [Ne ii] and [Ne iii] lines (Farrah et al. 2007). The ratio of the two probes the hardness of the radiation field, with starburst-dominated galaxies showing a line ratio of −1.3 ≲ log([Ne iii]/[Ne ii]) ≲ 0.1 in a sample of GOALS galaxies, though highly starbursting galaxies can show high ratios up to log ([Ne iii]/[Ne ii]) > −0.2 (Inami et al. 2013). We find log([Ne iii]/[Ne ii]) ∼−0.4 and ∼−0.8 for IRAS 20100−4156 and IRAS 03158+4227, respectively, which are consistent with the starburst range. Similarly, IRAS 20100−4156 and IRAS 03158+4227 show [O iii]/H$\beta \sim 1.2$ and [O iii]/H$\beta \,\sim 2.3$, respectively (Staveley-Smith et al. 1992; Meusinger et al. 2001), which are too low for AGNs but are consistent with LINER ratios based on the BPT diagnostic diagram (Kewley et al. 2006). Near-IR spectroscopy of the Paα and Brγ lines for IRAS 20100−4156 also does not show broad pedestals in the spectral lines, which would be evidence for emission from a broad-line region.

Overall, these diagnostics are consistent with the significant dust obscuration observed in the two sources, and we therefore cannot comment on the AGN luminosity based on any of these.

IRAS 20100−4156 and IRAS 03158+4227 are extremely dust-obscured sources, with high silicate optical depths (Spoon et al. 2013) and high nuclear column densities, as evidenced by the deep absorption in far-IR OH lines (González-Alfonso et al. 2017). This extreme obscuration renders most line diagnostics ineffective. This high dust obscuration also impacts the mid-IR decomposition, which is no longer applicable if the hot dust emission is completely obscured by cooler surrounding dust, as the warmer photons will be re-emitted at longer wavelengths. Another caveat for mid-IR decomposition is that the starburst template used for decomposition is derived using a combination of mid-IR spectra from starburst-dominated ULIRGs, which could always include some small AGN component in principle, possibly biasing the method toward finding higher starburst fractions. We note, however, that IRAS 08572+3915, an AGN-dominated highly obscured ULIRG (Efstathiou et al. 2014), is correctly identified as a highly AGN-dominated galaxy based on mid-IR decomposition, despite showing a deeper silicate optical depth (also no [Ne v] emission) and a geometry consistent with an edge-on torus, similar to the geometry for IRAS 20100−4156 and IRAS 03158+4227. This strongly suggests that a significant fraction of mid-IR photons can escape through optically thin lines of sight, despite high dust opacities. This scenario has been supported by simulations (e.g., Novak et al. 2012; Roth et al. 2012; Krumholz & Thompson 2013; Skinner & Ostriker 2015), although the results are sensitive to the numerical methods in question (Davis et al. 2014; Tsang & Milosavljević 2015). Overall, we do not find any evidence that the mid-IR decomposition is finding artificially low AGN fractions for our targets, though we cannot rule out more luminous and deeply buried AGNs with their dust torus completely obscured by cooler molecular gas.

Here we compare the inferred outflow momentum fluxes (${\dot{P}}_{\mathrm{OF}}$) to the expected momentum injection due to starburst and/or AGN activity, i.e., ${L}_{\mathrm{AGN}}/c$, ${L}_{* }/c$, ${L}_{\mathrm{IR}}/c$, and the predicted momentum injected from the sum of both (to first order, ${\dot{P}}_{\mathrm{sum}}\sim 20{L}_{\mathrm{AGN}}/c+3.5{L}_{* }/c$) for our sample of ULIRGs (Figure 11).

Zoom In Zoom Out Reset image size

We find that starburst activity alone is sufficient to provide the observed momentum boosts only in the weakest three outflows. We also find that nearly all the outflows show significant momentum boosts relative to the L_AGN, ${\dot{P}}_{\mathrm{OF}}\,\sim (5-50){L}_{\mathrm{AGN}}/c$. While these results are consistent with previous molecular gas outflow observations (e.g., Cicone et al. 2014; González-Alfonso et al. 2017), we also find significant differences. We find that high-momentum boosts up to ${\dot{P}}_{\mathrm{OF}}\sim 10{L}_{\mathrm{IR}}/c$ are not restricted to ULIRGs with f_AGN ≳ 0.5 but are seen in ULIRGs with a range of f_AGN ∼ 0.2–0.9. We detect no significant correlation between the outflow momentum flux and ${L}_{* }/c$ and ${L}_{\mathrm{IR}}/c$ (Spearman correlation coefficient $r({\dot{P}}_{\mathrm{OF}},{L}_{\mathrm{IR}}/c)\sim r({\dot{P}}_{\mathrm{OF}},{L}_{* }/c)\sim 0.3$; p-value of ∼0.5). We find a similar lack of correlation between the inferred outflow momentum flux and ${L}_{\mathrm{AGN}}/c$ (Spearman correlation coefficient $r({\dot{P}}_{\mathrm{OF}},{L}_{\mathrm{IR}}/c)\sim 0.4$; p-value of ∼0.3) and ${\dot{P}}_{\mathrm{sum}}$ (Spearman correlation coefficient $r({\dot{P}}_{\mathrm{OF}},{\dot{P}}_{\mathrm{sum}})\sim 0.4;$ p-value of ∼0.4).^{17 , 18}

The observed momentum boost for IRAS 03158+4227 ($\dot{P}\sim 9{L}_{\mathrm{AGN}}/c$) is consistent with the boosts predicted for AGN-driven outflows. For IRAS 20100−4156, on the other hand, we observe a high-momentum boost ($\dot{P}\sim 53{L}_{\mathrm{AGN}}/c$). The momentum boost is higher than expected in starburst-dominated outflow sources ($\dot{P}\sim 10{L}_{* }/c$ for IRAS 20100−4156, while $\dot{P}\sim 2-4\times {L}_{* }/c$ for starburst galaxies; Leitherer et al. 1999; Veilleux et al. 2005). Such high-momentum boosts can in principle be generated in deeply dust-obscured starburst nuclei (${\tau }_{\mathrm{IR}}\gg 1$) with high surface densities of star formation (${{\rm{\Sigma }}}_{\mathrm{SFR}}$), such that the star formation is Eddington-limited by radiation pressure on dust (${{\rm{\Sigma }}}_{\mathrm{SFR}}\,\sim 3000{M}_{\odot }$ yr⁻¹ kpc⁻²; Murray et al. 2005; Thompson et al. 2005). However, Roth et al. (2012) find that in such optically thick nuclei photons preferentially escape along optically thin lines of sight and limit the momentum boost to ∼5 L_IR/c. Furthermore, the high maximum outflow velocity in IRAS 20100−4156 is also unusual for starburst-driven outflows, although not impossible. Previous studies have consistently found significant correlations between the maximum outflow velocities in molecular gas outflows and L_AGN (Sturm et al. 2011; Spoon et al. 2013; Veilleux et al. 2013; Stone et al. 2016), suggesting that AGNs are responsible for the fastest-moving components of the outflow. Compact, Eddington-limited starburst activity has also been suggested to drive the high-velocity molecular outflow (∼1000 km s⁻¹) in a $z\sim 0.7$ galaxy (Geach et al. 2014), although the host galaxy is much less massive. While observations of outflows in starburst galaxies show that their maximum outflow velocities are strongly correlated with ${{\rm{\Sigma }}}_{\mathrm{SFR}}$, these probe ionized gas outflows, and the star formation surface densities are far below the Eddington limit (Heckman et al. 2015; Heckman & Borthakur 2016). For IRAS 20100−4156, assuming a radius of ≲0.2 kpc for the nuclear starburst and calculating the nuclear SFR to be ∼10⁻¹⁰ (L_*/L_⊙) (M_⊙ yr⁻¹), we find a surface density of star formation ${{\rm{\Sigma }}}_{\mathrm{SFR}}\gtrsim 3000{M}_{\odot }$ yr⁻¹ kpc⁻², which is above the Eddington limit. Overall, the outflow in IRAS 20100−4156 could in principle be driven purely by the starburst.

However, we consider this scenario unlikely based on the high outflow velocity observed. We instead favor the scenario that the outflow in IRAS 20100−4156 is driven by a combination of AGN and starburst activity, as neither AGN nor starburst feedback alone can provide the observed momentum boost, which we reiterate is a conservative estimate.

We compare the outflow energy flux (${\dot{E}}_{\mathrm{OF}}$) for all ULIRGs in our sample to L_AGN, ${L}_{* }$, L_IR, and the predicted energy flux from a combination of AGN and starburst activity (${\dot{E}}_{\mathrm{sum}}=0.05{L}_{\mathrm{AGN}}+0.02{L}_{* }$; Figure 12). We find that starburst activity can power the most massive outflows only if all the mechanical luminosity from SNe and stellar winds is converted into bulk motion of the gas with 100% efficiency, which is unrealistic. However, weaker outflows can be powered by stellar feedback even when realistic efficiencies are taken into account. Similarly, the outflows can be powered purely by AGNs (i.e., show energy fluxes ∼1%–3% L_AGN) only in ULIRGs with f_AGN ≳ 0.4, but not in starburst-dominated galaxies. We do not find that the outflow energy fluxes increase significantly with L_AGN and with ${\dot{E}}_{\mathrm{sum}}$ (Spearman correlation coefficients $r({\dot{E}}_{\mathrm{OF}},{L}_{\mathrm{AGN}})\sim 0.5$, $r({\dot{E}}_{\mathrm{OF}},{\dot{E}}_{\mathrm{sum}})\sim 0.5$; p-values of ∼0.3 and ∼0.2, respectively). We find a similar lack of correlation with L_* and L_IR (Spearman correlation coefficients $r({\dot{E}}_{\mathrm{OF}},{L}_{* })\sim 0.4$, $r({\dot{E}}_{\mathrm{OF}},{L}_{\mathrm{IR}})\sim 0.4$; p-value of ∼0.4 for both). We find that powerful outflows are detected in ULIRGs with f_AGN with a wide range of ∼0.2–0.9.

Zoom In Zoom Out Reset image size

We find that IRAS 03158+4227 shows an outflow energy flux of ∼1% L_AGN, which is close to the expected scaling. The outflow in IRAS 20100−4156, however, shows an outflow energy flux of ∼8% L_AGN, which is significantly higher than predicted from AGN activity alone. We therefore conclude that it is driven by a combination of the two, which is consistent with findings in the previous section.