DETECTIONS OF CO MOLECULAR GAS IN 24 μm BRIGHT ULIRGs AT z ∼ 2 IN THE SPITZER FIRST LOOK SURVEY^*

The nine targets observed by the PdBI were selected from a large sample of z ∼ 1–2 ULIRGs with Spitzer mid-IR spectra published in Yan et al. (2007). This parent sample consists of 52 sources initially selected with 24 μm flux density brighter than 0.9 mJy, and very red 24-to-8 μm and 24-to-R colors from the 4 deg² Spitzer Extragalactic First Look Survey (XFLS).⁶ The subsequently obtained Spitzer mid-IR spectra covering the rest-frame 4–20 μm determined that 74% of the sample is at 1.5 < z < 3.2, confirming the effectiveness of the initial color selection (Yan et al. 2004). The total IR luminosities (L_{3–1000 μm}) are in the range of 10^{12 − 13} L_☉. The complete mid-IR spectral and spectral energy distribution (SED) analysis shows that at least ∼75% of the sample contain dust obscured AGNs, which dominate the mid-IR 3–6 μm luminosities but star formation still contribute most of the far-IR luminosities (Sajina et al. 2007, 2008). Of the 52 sources, 44 galaxies were observed at 1.2 mm with the IRAM 30 m telescope using the 117 element version of the MAMBO array (Kreysa et al. 1998). Of these, seven are detected at ≳3σ (14 at ≳2σ), with an average rms ∼ 0.6 mJy. The 1.2 mm data of the full sample have been published in Lutz et al. (2005) and Sajina et al. (2008).

To measure cold molecular gas content and to quantify how it changes with mid-IR properties among these z ∼ 2 Spitzer ULIRGs, we obtained CO interferometric observations for sources with a broad range of mid-IR spectral properties, including sources with strong polycyclic aromatic hydrocarbon (PAH) emission, deep silicate absorption, and mid-IR power-law continua. The primary selection of our CO targets was the availability of accurate spectroscopic redshifts from Keck and GEMINI. This criterion has limited our targets to a small subset of the full sample. All of our targets have 1.2 mm observations from MAMBO. Of the nine sources, only one have 1.2 mm signal-to-noise ratio, S/N > 3σ, seven with S/N ∼1.81–2.58, and one with S/N < 1.5. Excluding the >3σ source, we stacked the remaining eight sources, yielding an averaged 1.2 mm flux of 1.06 ± 0.18 mJy. This value is higher than the averaged value of 0.5 ± 0.1 mJy for the 44 sources observed at 1.2 mm in the original Spitzer sample (52 objects; Sajina et al. 2008). This implies that our CO targets systematically have more far-IR emission, in comparison with the full Spitzer mid-IR spectroscopic sample at z ∼ 2.

Tables 1 and 2 list the coordinates, redshifts (from both CO and near-IR spectra) and the CO observational parameters for the nine sources observed by PdBI. Table 3 shows the ancillary observations from optical, near-IR, far-IR, and 20 cm photometry (see Sajina et al. 2008, for details). The near-IR spectra were obtained using NIRSPEC at the Keck and NIRI at the GEMINI. We have high-resolution H-band images from Hubble Space Telescope/NICMOS. This data set has been published in Dasyra et al. (2008) and is included in the analysis of a larger sample of z ∼ 0.3–2 bright 24 μm ULIRGs in M. Zamojski et al. (2010, in preparation).

Table 2. CO Observations

ID	Lines	νobs	Config.	Beama	P.A.	Timeb	Noise/Chan.	Chan. Width	Δc	rmsd
		(GHz)				(hr)	(mJy/chan)	(km s−1)	(km s−1)	(mJy beam−1)
MIPS429	(3–2)	107.642	C	48 × 389	68°	6	0.4	83	664	0.2
MIPS506	(3–2)	99.679	D	502 × 437	71°	4	0.35	75	525	0.25
MIPS8196	(3–2)	96.456	C	56 × 418	65°	4.5	0.39	62	...	...
MIPS8327	(3–2)	100.464	C	286 × 25	71°	9	0.7	60	540	0.12
MIPS8342	(2–1)	89.913	D	635 × 415	56°	9.5	0.6	67	670	0.1
MIPS15949	(3–2)	111.046	C	263 × 23	69°	10.1	0.4	54	756	0.12
MIPS16059	(3–2)	103.999	C	477 × 383	89°	5.5	0.5	87	522	0.25
MIPS16080	(3–2)	114.997	C	461 × 352	75°	10	0.7	72	360	0.35
MIPS16144	(3–2)	110.407	C	244 × 215	62°	5.4	0.5	81	810	0.13

Notes. ^aBeam is defined as the semimajor axis times the semiminor axis in arcseconds. ^bThis is the total on-target integration time, calculated equivalent to six antennas. ^cThe summed CO maps are integrated over a velocity width Δ with the center on the peak of the CO emission line. ^dThis noise is rms for the summed CO map made with the listed parameters.

Download table as:  ASCIITypeset image

The observations were carried out in 2007–2008 with the PdBI in D (two objects) and C (seven objects) configuration in order to maximize the sensitivity. The 3 mm receivers were tuned to the central observed frequency according to the near-IR spectroscopic redshifts. Of the total nine sources, we used the PdBI to observe the CO J(3–2) transition (ν_rest = 345.796 GHz) for eight objects and the J(2–1) transition (ν_rest = 230.538 GHz) for one object at redshift of 1.562. Typically, our sources were observed under good weather conditions with 1–2 tracks for a total of 5–10 hr, six antenna equivalent, on-target integration. Table 2 lists the observation parameters. The data reduction takes places in two stages using the IRAM GILDAS software. First, the raw data are calibrated using the software CLIC, developed at IRAM.⁷ Care is given to the phase, flux, and amplitude calibrations by rejecting any calibration anomalies due to meteorological conditions or electronics. After calibration, CLIC generates the visibility data, i.e., "uv" tables. Second, based on these uv tables, we extract CO maps and spectra using the IRAM MAPPING software.⁸ The synthesized, clean beam size is typically elongated, roughly (24 × 24) to (56 × 418) for the C configuration and (502 × 437) to (62 × 415) for the D configuration. The corresponding spatial resolution in linear size ranges from 20.5 to 53 kpc. For comparison, for the Spitzer 24 μm band, the full width at half-maximum (FWHM) of a point source is 6'' and the NICMOS H-band point-spread function (PSF) is 022.

Figure 1 shows the integrated CO maps, with the small insert at the left bottom of each panel indicating the beam size and positional angle. Before producing the summed CO maps, we rebinned the data with an original 10 MHz channel width by a factor of 3–4, yielding smoothed spectral data cubes with a channel width roughly 54–83 km s⁻¹. The CO maps shown in Figure 1 were made by summing over a number of velocity channels, ranging from 5 to 14 (equivalently, velocity width Δ ∼ 360–810 km s⁻¹ in Table 2), centered on the peak CO line emission. The final maps have 1σ noise values ranging from 0.1 to 0.35 mJy  beam⁻¹. Table 2 lists the noise parameters and the parameters used for making the summed CO maps. In Figure 1, the first CO contour starts roughly at 1σ with a step size of 1σ for all panels. The summed CO line emissions are detected at 4.7σ–9.7σ for eight sources. Table 4 lists the integrated CO line fluxes, velocity width over which the CO line integration is done, and derived CO luminosity and H₂ (+He) molecular masses.

Zoom In Zoom Out Reset image size

The CO line luminosity can be derived from the line intensity, following Solomon et al. (1997):

$$\begin{equation} L_{\rm CO} (L_\odot) = 1.04 \times 10^{-3} \Big ({I_{\rm CO} \over {\rm Jy\,km\,s}^{-1}}\Big) \Big ({\nu _{\rm obs} \over {\rm GHz}}\Big) \Big ({D_L \over {\rm Mpc}}\Big)^2 \end{equation} \tag{ 1 }$$

and

$$\begin{equation} L^{\prime}_{\rm CO} ({\rm K}\,{\rm km}/{\rm s}\,{\rm pc}^{2}) = 3.25 \times 10^7 {I_{\rm CO}} {\nu _{\rm obs}}^{-2} {D_L}^2 (1+z)^{-3}. \end{equation} \tag{ 2 }$$

Here L'_CO is the CO line luminosity in units of K km/s pc², I_CO is the integrated line intensity in Jy km s⁻¹, D_L is the luminosity distance in units of Mpc, and ν_obs is the observed frequency in GHz. The H₂ (+He) molecular gas mass can be estimated using the following equation:

$$\begin{equation} {M_{\rm gas} \over M_\odot } = \alpha \Big ({T_{{\rm CO}(3-2)}/T_{{\rm CO}(1-0)} \over 1}\Big) L^{\prime}_{\rm CO}. \end{equation} \tag{ 3 }$$

Here α is the CO J(1–0) luminosity to molecular gas mass conversion factor in units of M_☉(K km/s pc²)⁻¹ and T_CO is the effective brightness temperature of a CO transition. Here, we assume T_{CO(3 − 2)}/T_{CO(1 − 0)} = 1 and α = 0.8 M_☉(K km/s pc²)⁻¹ (Solomon et al. 1997; Downes & Solomon 1998), the same assumption used for SMGs and QSOs (Greve et al. 2005; Tacconi et al. 2006, 2008). Although a more realistic T_{CO(3 − 2)}/T_{CO(1 − 0)} ratio is probably between 0.5–0.7 (Wilson et al. 2008), we adopt our assumption in order to make direct comparison without additional corrections to the published SMG and QSO data.

To derive the total and far-IR luminosities (L_IR = L_{3–1000 μm}, L_FIR = L_{40–120 μm}), we fit the SED of our sources from near-IR to far-IR (see Table 3). The longest wavelength data available are MAMBO 1.2 mm, which is crucial for constraining the far-IR emission. All of the nine targets except one have weak 1.2 mm fluxes at 3σ or less. For these eight sources, the sigma weighted, averaged flux is 1.06 ± 0.18 mJy (5.9σ), providing constraining power in the SED fitting in Figure 2 (cyan, open stars). Our SED fitting uses three components: pure starburst template (the averaged SMG SED from Pope et al. 2006, 2008; green dot-dashed line), reddened type-I quasar SED (Richards et al. 2006; blue dashed line), and a 2 Gyr old single stellar population (SSP) model from Maraston (2005) to account for the near-IR emission (purple dot-dot-dashed line). The solid red line is the sum of these three components. The type-I quasar SED (blue) has to be reddened substantially (A_V ≳ 5) assuming a Milky Way (MW)-type extinction (Draine 2003). For most sources, this leads to a reasonably good fit to the mid-IR continuum and silicate absorption feature. However, two of our sources (MIPS8342 and MIPS15949) show red mid-IR continua but no significant silicate absorption. The above approach cannot fit this type of spectra. For these sources, we revert to our power-law continuum approach as used in Sajina et al. (2008).

Zoom In Zoom Out Reset image size

At the far-IR, four sources are detected in the 70 μm band, and only one at 160 μm. With so few far-IR detections, is the starburst-component really required? The only sources for which the answer is obviously yes are MIPS8342 (based on its 160 μm detection) and MIPS16144 (as it is a strong-PAH/mm bright source). For MIPS16080, its 70 μm detection is difficult to explain without such a component, unless we assume a somewhat cooler AGN template. In particular, the stacked 1.2 mm detection (1.06 ± 0.18 mJy) requires a minimum amount of star formation, as indicated by the quiescent spiral galaxy template shown as the solid pink line in Figure 2 (this is an Sd template from Polletta et al. 2007).

This naturally begs the question whether the starburst or the quiescent Sd template is best suited to fit the SEDs of our sources. The choice will greatly impact the calculated far-IR luminosity. We argue that in order to fit simultaneously the observed 70, 160 μm, and 1.2 mm fluxes, the starburst template is required. Sd templates can explain the stacked flux at 1.2 mm, but fail significantly with 70 μm detections and mid-IR spectra. The stacked limit at 70 μm is 2 ± 0.68 mJy, high enough that Sd template would fail to fit this limit by a large margin. Definitive solution to this problem will soon to be available from Herschel data with better spatial resolution and sensitivity. The luminosities derived from our SED fitting are tabulated in Table 5. We note that L_FIR is dominated by starburst component for all sources in this study.

Of the nine sources observed, eight yielded significant detections in the CO J(3–2) or J(2–1) transitions. The CO J(3–2) spectrum of MIPS16144 also shows a weak continuum at the level of 0.5 ± 0.1 mJy (λ_obs = 2.717 mm). This observed continuum flux is consistent with the full-IR SED (see the open triangle symbol in Figure 2). Figure 3 presents the CO maps, HST/NICMOS H-band images and mid-IR spectra, one row per source. The first column shows the CO contours overlaid on the 24 μm image (red; FWHM ∼ 6'') and the HST/NICMOS H-band image (dark; FWHM ∼ 022). All CO contours start at 1σ with an increment contour step size of 1σ. MIPS429 has no NICMOS H-band data, and we used the available WFPC2 F814W image instead. It has bright 24 μm flux but no optical counterpart in both ground-based R image and the HST/WFPC2 image. The second column shows the HST/NICMOS H-band images of our sources, providing morphologies at the rest-frame ∼5500 Å. The third and fourth columns are the CO and mid-IR spectra. The mid-IR spectra for MIPS16059 and MIPS16080 have higher S/N, and are from an ultra-deep Spitzer observation aiming to detect water ice and hydrocarbons among a sample of deeply embedded z ∼ 2 ULIRGs (Sajina et al. 2010).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The strength of PAH emission and the 3–6 μm hot dust continua are indicators of star formation and AGN, respectively. One interesting question is whether the mid-IR star formation indicator is correlated with cold dust and molecular gas. Figure 3 shows one obvious result, that is the detectability of CO gas is not strongly correlated with the mid-IR spectral properties, with the caveat that this finding is based on a small number of sources. Of the nine sources, seven have mid-IR spectra whose prominent features is the broad emission bump at 7–8 μm and/or fully or partially observed silicate absorption trough. This type of spectra is very characteristic for highly obscured 24 μm sources at z ∼ 2, as shown by several recent Spitzer observations (Houck et al. 2005; Yan et al. 2005, 2007). Local examples with this type of spectrum are IRAS F00183-7111 and NGC4418 (Spoon et al. 2004a, 2004b; Spoon et al. 2001). One characteristic of these deeply embedded local ULIRGs is their multi-temperature interstellar medium (ISM), ranging from hot gas near the central nuclei to cold gas to outer region (Spoon et al. 2004b). Our CO observations show that six of these seven sources with deep silicate absorption have cold molecular gas, one (MIPS8196) does not have CO detection at I_CO ≲ 3σ ∼ 0.5 Jy km s⁻¹.

The CO detection rate of our sample is high, likely due to the improved sensitivity and larger bandwidth of PdBI in recent years. Physically, such high detection rate implies that among 1 mJy or brighter 24 μm ULIRGs at z ∼ 2, a good fraction of them, including SB/AGN composites, deep silicate absorption systems, and AGN-dominated galaxies, could have abundant cold molecular gas. For these 24 μm bright ULIRGs with substantial black holes in their centers, one important yet unsolved question is how cold molecular gas is exactly turned into stars and feeds the growth of black holes. The answers will come from future high-resolution studies with ALMA.

Table 4 lists the integrated CO line intensity I_co in Jy km s⁻¹ and luminosity L'_CO for J(3–2) or J(2–1) transition. We emphasize again that to derive cold molecular gas mass, we assume T_co(3-2)/T_co(1-0) = 1, T_co(3-2)/T_co(1-0) = 1 and CO J(1–0) luminosity to molecular gas mass conversion factor α = 0.8 M_☉ (K km/s pc²)⁻¹. The eight sources have M_gas ∼ (1.02–3.02) × 10¹⁰ M_☉, with a median value of 1.7 × 10¹⁰ M_☉. In order to compare our sample with other galaxy populations at z ≳ 2 with CO detection, we compiled a list of SMGs based on the published data (Greve et al. 2005; Tacconi et al. 2006, 2008), shown in Table 6. This sample of 14 SMGs includes only sources with significant detections, and their median gas mass M_gas is 3.05 × 10¹⁰ M_☉. In addition, a sample of 19 QSOs and radio galaxies with low-resolution CO observations (Solomon & Vanden Bout 2005) is also used as a QSO comparison sample. As described in Section 3.3, we have also constructed the second QSO comparison sample, much smaller subset but with reliable dynamical masses from the recent high-resolution CO observations. Figure 4 (left panel) shows L'_CO versus redshift for the 24 μm ULIRGs and the comparison samples of SMGs, QSOs, and radio galaxies. The right panel in Figure 4 presents the histogram of L'_CO with the horizontal lines marking the median values of the three populations. The average CO mass of our 24 μm ULIRG sample is about a factor of 2 less than the average values calculated from the specific comparison samples of SMGs and gas-rich QSOs currently available in the literature. Kolmogorov–Smirnov tests on 24 μm ULIRGs versus SMGs and 24 μm ULIRGs versus QSOs derived the probabilities of two data sets drawn from the same parent sample of 0.19 and 0.189, respectively. We emphasize that the statistics is still too small to draw any definitive conclusions for the overall populations of these three types of sources.

Zoom In Zoom Out Reset image size

If far-IR emission is considered mostly from massive young stars born in recent starbursts, L_FIR can be directly related to star formation rate (SFR). Therefore, the ratio of L_FIR/L'_CO can be interpreted as describe (1) gas depletion timescale, τ_SF = M_gas/SFR or (2) star formation efficiency, SFE = L_FIR/M_gas, i.e., how much luminosity can be produced by 1 M_☉ of gas. If we take the SFR(M_☉ yr^-1) = ${c}*\bigl ({L_{\rm FIR} \over 10^{10}\;L_\odot }\bigr)$, where c = 0.8–3, depending on the definition of L_FIR (Kennicutt 1998; Meurer et al. 1997). We adopt c = 1.5 because we use L_FIR = L_{40–120 μm} and the commonly adopted relation by Kennicutt (1998) has c = 1.72 with L_FIR = L_{8–1000 μm}. The L_FIR is measured from the full SED fitting described in Section 2.3. The inferred SFRs are 32–787 M_☉ yr⁻¹, with the mean value of 585 M_☉ yr⁻¹. These values are substantially larger than normal star-forming galaxies, but a factor of 2–3 less than what has been observed for SMGs at the similar redshifts. Furthermore, if we assume that 70% of L_FIR from SMGs is due to star formation, as suggested by some recent studies (Alexander et al. 2008), the averaged SFR for SGMs would still be higher than that of 24 μm sources by a factor of 1.5–2. The inferred values for SFE and gas depletion timescale τ_SF range from 69–388 L_☉ M⁻¹_☉(〈SFE〉 = 330 L_☉ M⁻¹_☉) and 17–96 Myr (〈τ_SF〉 = 38 Myr), respectively. For comparison, star formation efficiency 〈SFE〉 is typically 180 ± 160 L_☉ M⁻¹_☉ for local ULRIGs (Solomon et al. 1997) and 450 ± 170 L_☉ M⁻¹_☉ for SMGs (Greve et al. 2005; Tacconi et al. 2006). And the gas depletion timescale τ_SF ranges from 10 to 100 Myr for SMG, and a factor of 10 longer for local LIRGs and ULIRGs (Solomon & Vanden Bout 2005). Figure 5 displays the L_FIR/L'_CO ratio versus L'_CO (or equivalently cold molecular gas), visually illustrating our conclusion that these bright 24 μm selected z ∼ 2 ULIRGs have less cold molecular gas and smaller SFE than that of SMGs.

Zoom In Zoom Out Reset image size

The CO velocity width (FWHM) can be measured by using either a single or double Gaussian fit or directly from the data (full width at the half peak intensity). Table 4 shows the results using both methods. Assuming a single component and without profile fitting (Table 4, FWHM2), we have velocity widths ranging from 128 km s⁻¹ to 824 km s⁻¹, with a median value of 406 km s⁻¹. If we use two velocity components for sources with double peak profiles, the averaged velocity width reduces to 275 km s⁻¹. With a single-component fitting, Figure 6 compares the CO velocity widths of our sources with that of SMGs and QSOs. Here, the SMG comparison sample is listed in Table 6 for 14 objects compiled from Greve et al. (2005), Tacconi et al. (2006), and Tacconi et al. (2008). The QSO comparison sample is compiled by Solomon & Vanden Bout (2005; also see Greve et al. 2005). The right panel presents the velocity distributions, with the horizontal lines marking the median values of 406, 635, and 350 km s⁻¹ for Spitzer z ∼ 2 ULIRGs, SMGs, and QSOs, respectively. As the local reference, the z ∼ 0 ULIRGs have a median CO velocity width of 240 km s⁻¹ (Sanders et al. 1991; Solomon et al. 1997). Kolmogorov–Smirnov tests on these three small samples produce the probabilities of 0.23 and 0.84, respectively, for ΔV(FWHM) of 24 μm ULIRGs versus SMGs and 24 μm ULIRGs versus QSOs drawn from the same parent distributions.

Zoom In Zoom Out Reset image size

Recent N-body/smoothed-particle hydrodynamic (SPH) numerical simulation combined with three-dimensional polychromatic radiative transfer by Narayanan et al. (2010) shows that SMGs are mostly produced by major mergers of gas-rich, star-forming-dominated progenitors, whereas 24 μm bright ULIRGs at z ∼ 2 could be formed by both major and minor mergers. This model predicts that CO velocity widths of bright 24 μm ULIRGs spread over 100–1500 km s⁻¹, overlapping with that observed among SMGs, but the median value is smaller than that of SMGs.

Spatial distribution of molecular gas holds the imprints of kinematic state of a galaxy. Although the majority of our sample is unresolved, two targets, MIPS506 and MIPS16144, show spatially extended CO emission across regions of ∼100 and 46 kpc, significant after taking into account of their beam sizes of 502 × 437 to 41 kpc × 36 kpc and 244 × 215 to 20 kpc × 18 kpc, respectively. MIPS506 shows two distinct components with a separation of 45 kpc. The CO spectra in Figure 3 show that MIPS506, MIPS16144, and MIPS8342 have double peaked velocity profiles, with single and double Gaussian fit (solid and dashed lines, respectively) to the observed data (histogram). A double peak CO velocity spectrum is indicative of either a rotating disks or of two galaxies in an ongoing merger. Our CO data measure 3 out of 9 (33% ± 19%) sources having double velocity components. The low-resolution CO data for the SMG comparison sample in Table 6 show 64% ± 21% (9/14) having double or multiple velocity components. If we consider pure Poisson statistical errors, these two fractions are consistent within 1σ.

One important question is whether the two velocity components in each of the three double peak sources are spatially separated. Figure 7 shows the CO contours overlaid on the 24 μm (pink) and HST H-band (black) images, with solid and dashed contours indicating two separate velocity peaks. Here, the integrated CO maps are summed over 4–5 velocity channels (see Table 2 for the channel width), covering the respective velocity peak. For example, the integrated CO maps for MIPS16144 shown in Figure 7 are summed from −730 km s⁻¹ to −330 km s⁻¹ for the velocity peak at −605 km s⁻¹ and from −330 km s⁻¹ to −70  km s⁻¹ for the velocity peak at −138 km s⁻¹. The similar maps are made for MIPS506 and MIPS8342, and shown in Figure 7. The rms σ values used for the contours are in Table 2. The two velocity peaks in MIPS16144 have a clear spatial separation of 13, which corresponds to 10.9 kpc. The two velocity components in MIPS8342 are spatially too close to be resolved. In the case of MIPS506, the separation is ∼55, corresponding to 45 kpc. Such a large distance, particularly for cold molecular gas, suggests that MIPS506 is probably a pair of merging galaxies, with substantial dust extinction in the rest-frame optical band. This explains the very faint multiple blobs (28–24 kpc separation) in the H-band image (6'' × 5'') shown in Figure 3. We have examined our shallow 3.5–8 μm IRAC images to see if the galaxy pair detected in the CO map is also detected in other bands, only an unresolved object is detected at 3.5 and 4.8 μm, and no detections at other two IRAC bands. The physical model for MIPS16144 and MIPS8342 is probably also merger, although rotating disk model is also consistent. Additional supporting evidence for Spitzer ULIRGs being mergers comes from the HST/NICMOS H-band images, with the detailed discussion in Section 3.4.

Zoom In Zoom Out Reset image size

For MIPS16144 and MIPS506 with spatially resolved CO maps, assuming a two-body interaction, we can approximate the total dynamical mass with M_dyn = R*v²_circ/G = R*ΔV²/(sin²(i)G), where sin(i) is the inclination angle, R is the separation between two merging objects, and ΔV is the observed CO peak velocity difference. R and ΔV are (10.9 kpc, 463 km s⁻¹) and (45 kpc, 174 km s⁻¹) for MIPS16144 and MIPS506, respectively. We derived M_dyn = 5.4 × 10¹¹ sin⁻²(i) and 3.2 × 10¹¹ sin⁻²(i) M_☉ for MIPS16144 and MIPS506. If assuming an averaged value of 〈sin²(i)〉 = 2/3, the gas fraction, M_gas/M_dyn, is 4% and 3% for MIPS16144 and MIPS506, respectively. Although several recent papers have published gas fractions for various types of galaxies, we caution that the equation used for calculating the dynamical masses based on observed emission line FWHM could differ by a factor of 2–5, see Neri et al. (2003), Erb et al. (2006), Tacconi et al. (2006), and Riechers et al. (2009).

The stellar half-light radii have been measured from the HST/NICMOS H-band (1.6 μm) images for our sample (Dasyra et al. 2008; M. Zamojski et al. 2010, in preparation). Using M_dyn = R_1/2*v²_circ/G = R_1/2*(ΔV^CO_FWHM)²/(sin²(i)G) and assuming that cold molecular gas disks are smaller than stellar half-light radii, we can also set limits on the dynamical masses of additional five sources in our sample. This assumption is usually correct for nearby galaxies (Sanders et al. 1991; Solomon et al. 1997). We note that our sources probably have fairly high dust extinction at the observed H band (rest-frame 5000 Å), which will cause systematic underestimates of the intrinsic stellar sizes. Using the half-light radii are listed in Table 5, we obtained M_dyn ∼ (0.2–2.2) × 10¹¹ M_☉ for MIPS8327, MIPS8432, MIPS15949, MIPS16059, and MIPS16080, using an averaged inclination angle. These rough estimates are on the same order of magnitude in comparison with the averaged dynamical mass of SMGs, 1.2 × 10¹¹ M_☉ (Greve et al. 2005; Tacconi et al. 2006).

The CO images and spectra in Figures 1 and 3 have revealed that at least one source, MIPS506, is a merger involving two galaxies with substantial cold molecular gas. In addition, MIPS8327 and MIPS16059 are likely interacting galaxies because their HST/NICMOS H-band images show close companions within separations of 3.3 and 4 kpc (04and 05), respectively. M. Zamojski et al. (2010, in preparation) performed detailed two-dimensional light profile fitting to the HST images, assuming a Sérsic profile model.⁹ Figure 8 compares the original H-band images with the model subtracted, residual images (reversed intensity scale) for five of our sources with CO detections, MIPS506, MIPS16144, MIPS8342, MIPS16080, and MIPS15949. Multiple blobs a clear tidal tail in Figure 8 suggests that MIPS506, MIPS16144, and MIPS16080 are probably the products of galaxy interactions. Although the original H-band images of MIPS8342 and MIPS15949 show isolated early type galaxies, but the faint disks/rings in the residual images indicate that they could be late stage mergers. Majority sources in our CO sample have merger signatures either in their CO maps or in the H-band images. Considering the published studies of SMGs based on HST images and integral spectroscopy (Chapman et al. 2003; Conselice et al. 2003; Swinbank et al. 2006), we speculate that SMGs, bright 24 μm ULIRGs, and gas-rich QSOs are connected by an evolutionary model involving mergers, with SMGs being in an early stage of mergers of two gas-rich, star-forming progenitors, and bright 24 μm ULIRGs with substantial obscured AGN in a later merger stage, while QSOs are fully merged with massive black holes in place, producing powerful feedback clearing away dust obscuration. This simple picture does not explain how SMGs and gas-rich QSOs could have such a different history of star formation and black hole growth. Specifically, most SMGs have very weak AGNs, and their estimated black hole to stellar mass ratio is roughly on the order of 1:10⁴ (Borys et al. 2005; Alexander et al. 2008), whereas, recent studies (see Table 7) found the same ratio for these z ≳ 3 QSOs is on the order of 1:10 (Riechers et al. 2008b, 2009). For our objects, black hole mass is calculated from the AGN bolometric luminosity estimated in the SED decomposition (see Figure 2). Stellar mass is computed from the rest-frame L_H assuming a M_stars/L_H ∼ 0.48 from a 2 Gyr SSP template of Maraston (2005). The inferred M_BH/M_stars is on the order of 1:10³. Again, this suggests that Spitzer ULIRGs may be a transitional type of sources between SMGs and QSOs. To really understand the physical relation among these three types of sources, we will need better UV optical spectra, complete SEDs, and high-resolution CO data for Spitzer ULIRGs.

Zoom In Zoom Out Reset image size