SSGSS: THE SPITZER–SDSS–GALEX SPECTROSCOPIC SURVEY

Interstellar dust absorbs from 50% to as much as 90% of the UV and optical starlight in star-forming galaxies (e.g., Calzetti et al. 1994; Wang & Heckman 1996; Buat et al. 2002; Sullivan et al. 2000; Bell & Kennicutt 2001; Calzetti 2001; Goldader et al. 2002), and the re-radiated IR photons constitute half of the bolometric luminosity in the local universe. Broad-band measures of the IR spectral energy distribution (SED) of galaxies have taken us a long way toward understanding the link between this emission and star formation; first with the Infrared Astronomical Satellite (IRAS; e.g., Helou 1986; Rowan-Robinson & Crawford 1989; Devereux & Young 1991; Sauvage & Thuan 1992; Buat & Xu 1996; Walterbos & Greenawalt 1996; Kennicutt 1998; Kewley et al. 2002), and subsequently with the Spitzer Space Telescope, with deep imaging surveys tracing obscured star formation in LIRGs, ULIRGs, and normal disk systems out to redshift 3 and beyond (e.g., Perez-Gonzalez et al. 2005).

However, several processes complicate the interpretation of broad-band measurements in the IR: the thermal IR continuum depends on both the temperature distribution and composition of the dust; aromatic bands in the MIR are linked to star formation, dust composition, and active galactic nucleus (AGN) activity, but are poorly understood; atomic and molecular lines span a wide range of ionizations and hence arise from a wide variety of environments; AGN and stellar emission compete with thermal dust emission in the continuum; and all of these may be affected by extinction (although more weakly than in optical and UV bands.) To begin to deconvolve the effects of these processes, and hence to unlock the diagnostic power of the IR spectrum of galaxies, high quality spectroscopy is necessary.

The Spitzer Infrared Spectrograph (IRS) has enabled great strides in this regard, especially for very luminous IR sources. We now have detailed MIR spectroscopic libraries of the most actively star-forming galaxies, both in the local universe (e.g., Brandl et al. 2006; Imanishi et al. 2010), and, increasingly, at higher redshifts (e.g., Yan et al. 2007; Murphy et al. 2009; Dasyra et al. 2009; Desai et al. 2009; Hernán-Caballero et al. 2009). The Spitzer Infrared Nearby Galaxies Survey (SINGS; Kennicutt et al. 2003) has been especially important in linking the primary components of the MIR spectrum with physical properties, including: linking the effects of metallicity, ionizing field, star formation, and H₂ intensity with polycyclic aromatic hydrocarbon (PAH) features (Smith et al. 2007; Roussel et al. 2007), physical modeling and characterization of dust content (Draine et al. 2007; Muñoz-Mateos et al. 2009a), and diagnosing abundances and the relative contributions of star formation versus AGN activity (Dale et al. 2009; Moustakas et al. 2010). With its local sample, one of the strengths of SINGS is its capacity for spatial resolution within its galaxies. This has enabled, for example, the mapping of the radial distributions of dust, gas, stars, and their evolution (Muñoz-Mateos et al. 2009a, 2009b, 2011) and the use of Hα attenuation as a tracer of obscured star formation (Prescott et al. 2007).

These results have provided a powerful insight into the inner workings of galaxies; however, it is challenging to draw general conclusions on the global properties of the SINGS sample due to its focus on the nuclear regions of these extended sources. In addition, SINGS is typical of other IRS samples, in that it is dominated by IR luminous galaxies. To characterize the IR spectra of a representative population of star-forming galaxies, we need increased depth, a greater redshift range, and more careful selection criteria.

SSGSS spectroscopy spans the 5–40 μm range, covering a wealth of diagnostic features. These include prominent emission bands from PAH molecules, a continuum fueled by reprocessed as well as direct starlight and by AGNs, and abundant emission lines revealing a very wide range of ionization states. Each of these features promises significant diagnostic potential.

2.2.1. PAH Bands

2.2.2. The Thermal Continuum

2.2.3. Emission Lines

The Lockman Hole, situated around α = 107, δ = 50°, is a ∼10 deg² field with one of the lowest column densities of interstellar material in the Milky Way, with $N_{{\rm H\,\mathsc{i}}} < 7\times 10^{19}$, approximately 5% of the Galactic median. This makes it ideal for low-surface brightness surveys, and so it has been the subject of extensive multi-wavelength surveys. These include deep ROSAT imaging, deep GALEX imaging (m_{FUV, NUV} 24.5 AB), SDSS imaging (m_r ∼ 22.2 AB) and spectroscopy, Deep Two Micron All Sky Survey (2MASS, × 6, m_K ∼ 17.8 AB; Beichman et al. 2003) and Spitzer/SWIRE (Lonsdale et al. 2003) IRAC (m ∼ 22.5, 20 for 3.6, 8.0 μm) surveys, as well as future programs (e.g., the UKIDSS deep survey over the full region). The SDSS primary spectroscopic sample targets all galaxies with r < 17.8 and yields ∼100 galaxies deg⁻¹.

The Lockman Hole sample of Johnson et al. (2006) forms the parent sample for SSGSS and consists of 916 galaxies at z < 0.2 with GALEX–SDSS–Spitzer/SWIRE all-band coverage. The wide spectral coverage of this sample—from FUV to FIR—allows a consistent treatment of dust absorption and the corresponding dust emission, providing a total accounting of the star formation, stellar mass, and dust attenuation for these galaxies. A detailed description of these and other derived measurements and physical parameters can be found in Section 8.2. SSGSS targets a subset of this Lockman Hole sample for observation with the Spitzer IRS. These galaxies are selected based on two brightness criteria:

• 1.  
  5.8 μm surface brightness. Galaxies fill the IRS Short-Low slit, and so at short wavelengths we apply a surface brightness limit of I_5.8 μm > 0.75 MJy sr⁻¹, based on IRAC 5.8 μm flux and half-light diameter (the median/mean half-light diameter is 5 Å for the entire sample).
• 2.  
  24 μm flux. At long wavelengths most of the galaxies are unresolved, and so a flux limit of F_24 μm > 1.5 mJy is applied.

The above criteria yield 154 galaxies. We further restrict the sample by performing an even sampling of the planes of near-UV (NUV)-3.6 μm versus stellar mass (from Brinchmann et al. 2004) and NUV-3.6 μm versus D_n(4000); in the dense regions of these spaces, we select the galaxy with the highest 24 μm flux. This process ensures that the sample spans a comprehensive range of physical properties. Figure 1 shows the distributions of NUV-3.6 μm color versus stellar mass and D_n(4000) for the SSGSS galaxies compared to the Lockman Hole sample, and illustrates that the two span a similar range of galaxy properties. The Lockman Hole sample is representative of the subsample of SDSS galaxies that have GALEX FUV detections; that is to say, normal, star-forming galaxies, consisting of both early- and late-type,s with 1.1 ≲ D_n(4000) ≲ 2.1.

Zoom In Zoom Out Reset image size

The sample observed with Spitzer consists of 101 galaxies. Due to a fault with the SL spectrum for SSGSS 40, the current SSGSS data release consists of 100 galaxies. We preserve the original indexing for consistency.

Within the SSGSS sample, we designate the bright sample and the faint sample, with the bright sample consisting of the 33 galaxies with F_24 μm > 5 mJy. The entire sample is the subject of low-resolution 5 μm–40 μm IRS spectroscopy, while the bright sample is also observed with 10–20 μm high-resolution spectroscopy.

This final SSGSS sample has a redshift range of 0.03 < z < 0.2 with median redshift 0.085, and a total infrared luminosity (L_TIR) range of 3.7 × 10⁹ L_☉ < L_TIR < 3.2 × 10¹¹ L_☉, with median 3.9 × 10¹⁰ L_☉. The bright sample has a redshift range of 0.04 < z < 0.2 with median redshift 0.077, and an L_TIR range of 5.1 × 10⁹ L_☉ < L_TIR < 1.9 × 10¹⁰ L_☉ with median 6.9 × 10¹⁰ L_☉. The flux and surface brightness cuts do eliminate galaxies with very low stellar masses (2 × 10⁹ M_☉ ⩽ M ⩽ 2 × 10¹¹ M_☉), metallicities (8.7 ⩽ log (O/H) + 12 ⩽ 9.2), and extinctions (0.4 < A_Hα < 2.3).

Table 1 details the 101 galaxies in the original SSGSS sample, and Figure 2 shows the distribution of redshift, stellar mass, L_TIR, and D_n(4000) of the sample.

Zoom In Zoom Out Reset image size

The full SSGSS sample was observed with the IRS on board the Spitzer Space Telescope as a Cycle-3 Legacy Survey. All spectroscopic observations were performed in spectral mapping mode. Observations were divided between 29 Astronomical Observation Requests (AORs), each consisting of four to eight galaxies observed in cluster mode. All galaxies within an AOR were observed consecutively, using guided telescope shifts to move between targets.

The entire SSGSS sample was observed with the IRS "low-res" Short-Low (SL) and Long-Low (LL) modules. The SL module spans 5.2–14.5 μm with resolving power R = 60–125 and has a slit width of 36–37. The LL module spans 14–38 μm with resolving power R = 57–126 and has a slit width of 105–107. Both the SL and LL modules are divided into two sub-slits—the first and second orders—abbreviated SL1 (7.4–15.4 μm), SL1 (5.2–8.7 μm), LL1 (19.5–38 μm), and LL2 (14.0–21.3 μm). The source is positioned on opposite halves of the detector for the first and second orders. A third "bonus" order from an additional sub-slit is obtained with the second order.

Four offsets in the direction of the slit were used to produce a one-dimensional dither pattern, with two exposures at each position. This strategy has been shown to increase the quality of output spectra, in particular for the handling of rogue pixels, undersampling, and flat-fielding as described in the IRS instrument support team document Obtaining Spectra of Very Faint Sources with the IRS, and in Teplitz et al. (2007).

For low-res spectroscopy our sensitivity target was 5σ per resolution element for the continuum and PAH band features, after moderate smoothing for the faintest sources. To achieve this sensitivity, our bright-sample exposure times are 8 minutes in both the SL and LL modules and our faint-sample exposure times are 8 minutes in the SL and 16 minutes in the LL module.

The 33 galaxies in the SSGSS bright sample were also observed with the IRS "hi-res" Short-High (SH) module. This module spans 9.9–19.6 μm with R ≈ 600, and has slit width 47.

For the hi-res spectroscopy our target line flux was 1 × 10⁻¹⁸ W m⁻² at 5σ, based on the line fluxes for [Ne ii], [Ne iii], [Ar ii], and [S iii] for star-forming galaxies with infrared luminosities comparable to those in the bright sample. Our hi-res exposure times are 16 minutes, split over separate exposures.

For each hi-res AOR, eight separate sky spectra with a total of 16 minute exposure times were also taken.

A crucial aspect of this program is the comparison of MIR spectral properties to SDSS optical properties and so it was essential to ensure overlap between the slit and the 3'' SDSS fiber. All slits are wider than the SDSS aperture, and so the SDSS aperture was centered within each slit with at least 05 accuracy. Acquisition ("peak-up") imaging at each AOR enabled us to achieve this accuracy, as described in Section 4.5.

Slit orientations were not specified and so are random. As the SL and LL observations are taken consecutively, the relative orientation of these slits on each galaxy is the same as their relative orientation on the spacecraft (859). SH slit positions are randomly oriented with respect to the low-res positions.

To achieve the accuracy needed for our slit positioning, we preceded the observations in each AOR with a pointing-mode peak-up image. The IRS peak-up facility uses centroid positioning from an image of the field in the SL detector array to position the science target in the slit. In our case, we used bright sources with known offsets to the targets. Despite the high galactic latitude of our sources, there were sufficient bright, low-proper-motion sources in each field to ensure quality peak-ups.

In addition, each galaxy was observed with dedicated 4 × 30 second peak-up imaging. These peak-ups utilized the blue filter, with spectral coverage of 13.3–18.7 μm. This ∼16 μm coverage gives us photometric data intermediate to the IRAC 8 μm and MIPS 24 μm imaging.

Figure 18 (Appendix A) shows the dedicated peak-up images for all galaxies in the sample, including slit positioning.

A solar flare during one AOR resulted in very high cosmic ray counts for five targets. The affected galaxies are SSGSS 20, 22, 23, 24, and 25.

Standard IRS processing was performed by the Spitzer Pipeline version S15.3.0. Pipeline steps included ramp fitting, dark subtraction, droop, linearity correction and distortion correction, flat fielding, masking and interpolation, and wavelength calibration.

No separate sky images were taken with the low-res modules and so sky frames were constructed from combined science frames. Between the first- and second-order exposures, the source is shifted to opposite sides of the detector. This results in a sky exposure for the order not currently containing the source. Within each AOR, all such sky exposures were median combined to produce a master sky frame for each order.

Bad and rogue pixel masks for each exposure were created with IRSCLEAN (version 1.9), with AGGRESSIVE keyword set to 2, and with negative pixels also masked. We also applied the campaign rogue pixel masks released by Spitzer.

Figure 3 shows example frames for the SL module through the reduction process, including the initial post-pipeline frame with exposures combined, the sky-subtracted frame, and the cleaned frame.

Zoom In Zoom Out Reset image size

Extraction of one-dimensional spectra was performed using SPICE version 2.0.1. The default setting was used to create a wavelength collapsed spatial profile, and the center of the profile was determined manually due to excess failures of the auto-center option. SPICE then performed spectral extraction along this peak. The point source tune module applied flux corrections for changes in the wavelength-dependent point-spread function (PSF) width and aperture size, and assumed emission by a point source.

5.1.1. Low-res Combining and Stitching

The extracted one-dimensional spectra were combined by weighted mean. First, the spectra at the four dither positions along the slit were combined for each order. Then the first, second, and bonus orders were stitched to produce separate SL and LL spectra using the following semi-manual process: by default, the central 50% of each overlap region was combined by weighted mean and the outer quartiles of the overlap region were set to the nearest order. In select cases the combined region was adjusted to avoid obvious drops in transmission at the ends of either order. The SL and LL modules were combined by a clean cut between the spectra, determined manually to avoid end-of-channel transmission drops. Figure 4 illustrates the method used for stitching spectra.

Zoom In Zoom Out Reset image size

The overlap region of the SL and LL modules is ∼14–14.5 μm, and in this region the slit widths are 37 and 105 for SL and LL, respectively. The approximate LL-to-SL ratio of admitted light from centrally aligned point source in the overlap region is ∼1.3, based on a simple model of the Spitzer PSF (Smith 2004), and the overall LL sensitivity is a factor of ∼5 higher than the SL module in this region (Spitzer Observer's Manual, p.160). While these differences are accounted for by the pipeline reduction assuming a centrally aligned point source, for extended sources we may expect an offset at their intersection when the spectra are stitched.

In fact no galaxies exhibit noticeable discontinuities across the modules in this region, indicating that deviations from the point-source approximation in the overlap region are small, even for the SL module at ∼14–15 μm. No re-scaling was necessary or performed.

As with the low-resolution spectroscopy (Section 5.1), the Spitzer Pipeline version S15.3.0 was used to perform standard IRS processing. Unlike the low-res data, for the hi-res observations we combined the eight separate exposures of each galaxy before extraction. This improved the signal to allow higher quality extractions.

As noted in Section 4.3, separate hi-res sky frames were taken for each AOR, median combined from eight 16 minute exposures. Masking followed the same method used for the low-res spectroscopy, utilizing IRSCLEAN (version 1.9) and the Spitzer campaign rogue pixel mask. Likewise, extraction was performed as for the low-res data, using SPICE v.2.0.1. The hi-res orders were stitched into a single spectrum using a similar method to the low-res spectra: order-overlap regions were combined by taking the weighted mean within the central 50% or the overlap regions, while for the outer quartiles the nearest order was used.

Figure 5 shows example frames for the SH module through the reduction process, including the initial post-pipeline frame with exposures combined, the sky-subtracted frame, and the cleaned frame.

Zoom In Zoom Out Reset image size

We estimate the continuum sensitivity achieved by determining the signal-to-noise ratio (S/N) in the processed spectra at relatively featureless regions of the spectra. We compare the mean flux to the corresponding noise in the regions 9–10 μm and 20–24 μm for the low-res spectra and 14.7–15.3 μm for the hi-res spectra. The noise is taken to be the standard deviation in these regions after subtraction of a power-law fit. Figure 6 shows the continuum S/N for low-res and hi-res spectra versus IRAC 7.8 μm and MIPS 24 μm mag. At ∼22 μm nearly all sources exceed the target S/N = 5, with a minimum S/N of 4, and have the expected tight relationship with 24 μm mag. At ∼10 μm the variable silicate absorption results in a number of sources with S/N < 5. In this case there is greater variability of S/N with photometric magnitude due to this absorption and the overlap of more PAH bands with the IRAC 7.8 μm band. 22 μm S/N is provided in Table 1.

Zoom In Zoom Out Reset image size

We estimate the S/N for the [Ne ii]_12.8 and [Ne iii]_15.6 emission lines in the hi-res spectra by comparing the integrated flux of the lines to the standard deviation in the relatively smooth, blueward regions at 12.25–12.75 μm and 15–15.5 μm. Figure 7 shows line S/N versus the line flux (see Section 6.3). [Ne ii] and [Ne iii] are detected in all galaxies with S/N > 5. In most cases, both [Ne ii] and [Ne iii] are detected with S/N > 10. There is strong scatter in the trend between line flux and S/N due to the weakness of the trend between continuum and line luminosity, and this is especially pronounced in [Ne iii], where the line emission may be linked more to an AGN component than to star formation.

Zoom In Zoom Out Reset image size

The flux uncertainties reported by SPICE, which are included in the released data, have been checked for the low-res modules by comparing the multiple dithered observations for each galaxy. These uncertainties were found to be accurate to within a factor of 1.5.

An increasing fraction of the wavelength-dependent Spitzer PSF falls outside a given IRS slit with increasing wavelength. The standard spectral extraction with SPICE corrects for this slit loss assuming emission by a point source. At > 16 μm the combination of the larger Spitzer PSF and the wider IRS slits means that all galaxies are well approximated as point sources (Figure 18, Appendix A). However, a number of the SSGSS galaxies have sufficient spatial extent that they are not well approximated by point sources in the narrower SL and SH modules.

We investigate slit loss in the low resolution modules by comparing synthetic photometry from the reduced, calibrated low-res spectra to large-aperture photometry obtained from SWIRE imaging (IRAC channel 4 7.8 μm and MIPS 24 μm), and from the SSGSS 16 μm peak-up imaging. Synthetic photometry is obtained by integrating the spectra using k_correct (v4.14; Blanton03b). Figure 8 shows the difference in these magnitudes as a function of the r' band Petrosian radius (R_P), with aperture photometry performed in 7'' and 12'' radius apertures for 8 μm and 24 μm, respectively, as described in Johnson et al. (2007). Differences may be attributed to a number of sources (e.g., source confusion in the imaging aperture photometry, relative calibration errors) but any trend with the angular size of the galaxy is suggestive of slit losses. No such trend is observed at 16 μm and 24 μm. However, at 8 μm a significant trend is apparent, indicating slit losses in the SL module.

Zoom In Zoom Out Reset image size

To isolate the effect of slit loss from any calibration uncertainties, we perform photometry of the 8 μm IRAC images within the region corresponding to the SL slit aperture (a 365 × 96 rectangle), and compare this to the synthetic and circular aperture magnitudes. Figure 9 (upper panel) shows the difference between slit aperture and synthetic magnitudes. A 0.3 mag aperture correction is applied by SPICE at 8 μm and so the synthetic magnitudes are higher than the derived slit aperture magnitudes. Taking this offset into account, the slit aperture samples more flux than the extracted spectra, with an average offset of ∼0.2 mag and similar scatter. The magnitude difference is independent of galaxy size, indicating that the trend observed in Figure 8 does not result from calibration effects. Figure 9 (lower panel) shows slit aperture versus circular aperture magnitudes, revealing a similar trend to that observed with the 8 μm synthetic magnitudes. The maximum difference is 1–1.5 mag for R_p(r') ≳ 9'',

Zoom In Zoom Out Reset image size

The apparent slit losses in the SL module mean that care is necessary when considering the shape of the (≲ 16 μm) SED of large galaxies. The difference between the synthetic and photometric magnitudes at 8 μm should provide a reasonable estimate of the aperture correction. These corrections are provided Table 1.

Figure 19 (Appendix B) shows the reduced, combined, and stitched SL and LL spectra for the 100 galaxies in the current data release. For each galaxy the SSGSS catalog number, BPT type (Baldwin et al. 1981; Kewley et al. 2001; Kauffmann et al. 2003a), redshift, stellar mass, NUV-R color, and D_n(4000) are given.

The relatively low sensitivity of the 5.25–7.58 μm SL2 module, coupled with the relative faintness of many galaxies in this range, results in very low SL2 signal in 10 sources. This spectral region should be used with caution in these sources, which are noted in Table 1.

For one galaxy, SSGSS 56, the SL–LL overlap region falls directly on the redshifted 12 μm PAH complex and the [Ne ii]_12.8 line. End-of-order transmission drops in both the SL and LL modules affect these features, and so both are lost in this spectrum.

Figure 20 (Appendix B) shows the reduced, combined, and stitched SH spectra for the 33 galaxies in the SSGSS bright sample. For each galaxy the SSGSS catalog number, BPT type, redshift, stellar mass, NUV-R color, and D_n(4000) are given.

We measure emission line strengths by integrating line regions after subtraction of the continuum. Here, the continuum refers to all non-emission line features, including PAH features and the thermal continuum. The continuum is determined using the PAHFIT code of Smith et al. (2007), which fits a physically motivated model to each spectrum using χ² minimization. We describe our use of PAHFIT in detail in O'Dowd et al. (2009; hereafter OD09). When applied to the hi-res spectra, we remove emission line fitting from the process and mask the emission line regions. We are not concerned with accurate measurement of continuum features—only that we find a reasonable extrapolation of the continuum in the region of the emission lines.

The emission line strengths measured by PAHFIT from the hi-res spectroscopy are highly reliable, as the continuum near most lines is reasonably smooth, allowing robust subtraction. These line strengths compare closely to those measured using a simple linear fit to the continuum surrounding each line. In the case of the [Ne ii]_12.8 line, PAHFIT is significantly more reliable than cruder continuum-fitting methods because of the close proximity of this line to the peak of the 12.7 μm PAH feature.

Figure 10 compares [Ne ii]_12.8 and [Ne iii]_15.6 line strengths between the low- and hi-res spectra. Although there is general agreement, on average the low-res spectroscopy shows weaker line strengths by ∼20%–50%, with similar scatter compared to the hi-res spectra. In four galaxies, PAHFIT claims detection of [Ne iii] lines in the low-res spectra where they are undetected at high resolution. The uncertainties reported by PAHFIT for the low-res [Ne ii] lines are significantly underestimated, suggesting that there is more degeneracy between this line and the 12.6 μm PAH complex than estimated by the fit.

Zoom In Zoom Out Reset image size

Table 2 gives emission line strengths measured from the hi-res spectra. Line strengths from the low-res spectra can be found on the SSGSS Web site (see Section 8.2).

Figure 11 (upper panel) shows the composite of the low-res SL and LL spectra for the entire SSGSS sample, normalized between 9 and 11 μm. Primary PAH features are prominent, including a pronounced 17 μm complex. Several emission lines are also distinct.

Zoom In Zoom Out Reset image size

Figure 11 (lower panel) shows the composite of the SH spectra for the SSGSS bright sample, normalized between 14 and 15 μm. PAH features have significantly more detailed line profiles than in the low-res spectra, including clear resolution of the 16.5 μm feature from the 17 μm complex and interesting structure in both the 11.3 μm and 12.7 μm features. In addition, the [Ne iii]_15.6 μm line is well resolved from the 12.7 μm feature.

The SSGSS sample spans the range of physical properties of "normal" star-forming galaxies, as described in Section 3. We investigate the difference between the composite SSGSS spectra and galaxy spectra of previous IR surveys, which have typically been dominated by more extreme starburst galaxies.

Figure 12 shows the low-res SSGSS composite spectrum (normalized between 9 and 11 μm for composite construction) overplotted with an Infrared Space Observatory (ISO) spectrum of M82 (Sturm et al. 2000) and the Spitzer local starburst composite of Brandl et al. (2006). These spectra are normalized according to the results of the fits described below. Also plotted is the 0.8 < z < 1.2 LIRG composite of Dasyra et al. (2009), normalized at the blue end of this spectrum, 6.5–6.8 μm.

Zoom In Zoom Out Reset image size

M82 and the local starburst composite have similar EWs for PAH features at 11.3 μm and blueward. However, the EW of the 12.7 μm feature is lower in M82 and the EW of the 17 μm feature is lower in both. These starburst galaxies appear to have stronger long wavelength continuum emission than the galaxies of the SSGSS sample. Note that the MIR spectral shape of M82 is within the range seen for the Brandl et al. (2006) starbursts. The composite for z ∼ 2 ULIRGs shows significantly reduced EWs for all PAH bands, and a thermal dust component slope consistent with that of the local starburst template; however, the short wavelength range of the ULIRG composite makes this difficult to establish from these data.

To determine whether the difference in spectral shape between local starbursts and SSGSS galaxies is consistent with a difference in small-grain dust temperature distribution, we fit the M82 and local starburst composite spectra with a model consisting of the SSGSS low-res composite spectrum plus a dual-temperature blackbody dust component, in the case of M82 masking a 10 μm window around the 9.7 μm silicate absorption feature. Both are well fit with the addition of a 120 K component plus a single cooler component at around 80 K, and are poorly fit with components warmer than 130 K. The lower EWs of the 12.7 μm and 17 μm features are reproduced in the fit to the starburst composite, while only the 12.7 μm feature is accurately reproduced in M82; the 17 μm feature has lower EW than can be explained by a heightened continuum. In OD09 it was seen that the 17 μm feature is relatively weak compared to the 7.7 μm feature in galaxies undergoing more intense star formation. This is consistent with our observation of the weaker feature in M82. The results of these fits are also shown in Figure 12. The relative normalization of all spectra in this figure are determined by the best fits.

If we assume that the emission between 20 μm and 30 μm in the SSGSS composite arises solely from a 120 K thermal dust continuum component, then we estimate the additional flux from this component as a factor of two higher in the local starburst composite and a factor of eight higher in M82 compared to the SSGSS average. Although the additional continuum light apparent at the longer wavelengths of the starburst spectra is likely to be due to dust at a range of temperatures, this simple model shows that the difference in spectra is consistent with starburst galaxies having a higher proportion of emission from ≲120 K dust than more quiescent galaxies.

The shape of the MIR spectra of the most extreme IR emitters tends to be dominated by either a bright continuum—from either AGNs or hot dust—or by heavy silicate absorption centered at ∼9.7 μm. Spoon et al. (2007; hereafter S07) showed that this gives rise to a bimodality in the space of 6.2 μm EW versus silicate absorption strength for ULIRGs and other IR-luminous galaxies and for AGNs, from analysis of sources observed with Spitzer and ISO.

Figure 13 shows the location of the SSGSS galaxies in this space, with the regions occupied by the S07 sources shaded. We calculate our indicator of absorption strength similarly to the method used by S07 for starburst galaxies. We take the ratio of the 9.5–10.5 μm minimum to the "unabsorbed" flux at this point, the latter determined by interpolation using a linear fit between 4.5–5 μm and 14.5–15 μm regions, which are relatively unaffected by PAH emission or silicate absorption. 6.2 μm PAH EW is calculated using PAHFIT to measure the feature strength, as described in Treyer et al. (2010; hereafter T10). This method differs from that used by S07, who determine the continuum, and hence the EW, by fitting a spline between the edges of the 6.2 μm feature.

Zoom In Zoom Out Reset image size

Methods for EW determination based on estimation of the continuum from the boundaries of the 6.2 μm feature result in saturation at widths greater than ∼0.5–1 μm. This is due to the fact that, at higher line strengths, the continuum estimation includes increasing amounts of light from the 6.2 μm feature and the neighboring 7.7 μm PAH complex in the continuum estimation. At EW ≳ 0.5–1 μm the flux at the points typically used to estimate the continuum is in fact dominated by PAH emission. Sargsyan & Weedman (2009) report 6.2 μm EWs for SSGSS galaxies, finding EW < 1 μm for the entire sample. They calculate EW using a continuum fit anchored at 5.5 μm and 6.9 μm. This approach, like the spline method, is vulnerable to underestimates and saturation at high EWs due to the presence of significant PAH flux in the regions used for continuum estimates. T10 presents an analysis of this effect, and a comparison with the Sargsyan & Weedman (2009) results. As many SSGSS galaxies have strong 6.2 μm features, it is important that their EWs be measured using full continuum+PAH profile fitting.

In Figure 13 we can see that SSGSS galaxies have silicate absorption indicators consistent with the low absorption locus. A number of our sources are consistent with the starburst galaxies in S07's analysis, but much of our sample extends to significantly higher EWs, in part because of the EW saturation issue. Sources with AGN components by BPT designation (Baldwin et al. 1981; Kewley et al. 2001; Kauffmann et al. 2003a) have brighter continua, and hence lower EWs, than purely SF galaxies. These EWs may also have been reduced by destruction of PAH by the AGN.

For star-forming galaxies, total infrared luminosity (L_TIR) is a good proxy for the rate of current star formation, with much of the starlight in dusty starburst environments being reprocessed into the IR. SSGSS has already shown that PAH emission is also strongly tied to a galaxy's star formation history; the relative intensities of PAH bands trace specific star formation indicators (OD09), while the luminosities of individual PAH bands correlate strongly with total SFR (T10).

Understanding this latter relation, and in general the relative strength of the PAH contribution to galaxies' MIR emission, is important for modeling the IR SEDs of higher redshift galaxies where the PAH bands cannot be so clearly resolved, and for SED modeling in large photometric surveys such as SWIRE (Lonsdale et al. 2003). T10 demonstrated the near-linearity of the relation between the intensity of individual PAH features and both L_TIR and SFR. For the purpose of model fitting to MIR data in which individual PAH features are not well resolved, it is also valuable to determine the link between total PAH luminosity (L_PAH) and SFR and L_TIR.

We measure L_PAH using PAHFIT (Smith et al. 2007; see Section 6.3), summing the integrated strengths of all PAH bands between, and including, the 6.2 μm to 17 μm features. The relatively weak 3.6 μm feature falls outside our spectral range and is not included in our determination of L_PAH. We correct L_PAH for aperture effects, comparing SWIRE and SSGSS peak-up photometry to synthetic photometry (see Section 5.4), and interpolating corrections to PAH peak wavelengths. L_TIR is determined by fitting Draine & Li (2001) model SEDs to the Spitzer (IRAC, IRS blue peak-up, and MIPS) photometry, and then integrating between 3 and 1100 μm. The SFR is from Brinchmann et al. (2004), who fit the models of Charlot & Longhetti (2001) to strong emission lines of galaxies in the SDSS spectroscopic sample, using a Kroupa initial mass function (Kroupa 2001) and the Charlot & Fall (2000) dust model. We use the aperture-corrected median of the derived SFR likelihood distributions, designated SFR_e.

Figure 14 (left) shows L_PAH versus SFR_e. These quantities are strongly correlated for the full SSGSS sample; however, we are interested in calibrating L_PAH to star formation, and so we perform regression analysis for SF galaxies only:

$$\begin{eqnarray} \log _{10} L_{{\rm PAH}}(L_{\odot })&=&[0.99\pm 0.05] \log _{10} {\rm SFR}_e(M_{\odot }\,{\rm yr^{-1}})\nonumber\\ && + [8.8 \pm 0.1]. \end{eqnarray} \tag{ 1 }$$

Zoom In Zoom Out Reset image size

The slope is consistent with that presented in T10 for the 7.7 μm complex. As also found by T10, galaxies with AGN components generally fall on the same locus as purely star-forming galaxies, although they have significantly greater scatter. This is primarily due to the fact that SFR is determined with much lower accuracy for the more evolved populations in which BPT-designated AGNs are exclusively found.

This is further explored in Figure 14 (middle panel), which highlights the youngest 50% of the SSGSS sample, as defined by specific SFR_e (SSFR; SSFR >10⁻¹⁰ yr⁻¹), chosen simply to divide the sample in two. The correlation is significantly tighter when only these more actively star-forming galaxies are included. Figure 14 (right panel) shows the standard deviation in the log residuals for L_PAH after subtraction of the best linear fit for a range of subsample sizes. The subsamples are sorted to contain the youngest stellar populations, selected by three different measures: SSFR, D_n(4000), and Hα EW. In all cases, scatter decreases as the dominance of the young population increases—almost monotonically for D_n(4000) and Hα EW, and for these measures the scatter flattens significantly for the subsample containing the youngest ∼60% of sources, and for smaller "youngest" subsamples. The diagnostics corresponding to the ∼50th–60th percentile transition are D_n(4000) = 1.4, Hα EW = 18 Å, and SSFR = 10⁻¹⁰ yr⁻¹. For more evolved sources, the exceptionally tight trend between log L_PAH and log SFR_e (σ < 0.2) gains significant scatter. Again, this is primarily due to the inaccuracy with which SFR is measured for evolved populations.

Figure 15 (upper panels) shows the mass-weighted relationships. The ratio of L_PAH to stellar mass (M_* is plotted versus optical star formation diagnostics from SDSS: the narrow band 4000 Å break measure D_n(4000) (Kauffmann et al. 2003b) and Hα EW (left and middle panels, respectively), and also SSFR (right panels), which is described above. There are tight linear correlations between all quantities, although again we see the increased scatter in SFR for evolved galaxies. The best-fit regression lines for the SF galaxies are also shown. The mean 1σ residual scatter in L_PAH/M_* for SF galaxies after subtraction of the best-fit lines is ∼20% for all three cases. For D_n(4000) and Hα EW, the residual is ∼20% even including galaxies with AGN components, indicating that these age diagnostics are good predictors of mass-weighted PAH intensity for most galaxies. This emphasizes that the increased scatter seen in L_PAH and L_TIR versus SFR in evolved populations is due to inaccuracy in measuring SFR rather than, say, a blurring due to evolved stars contributing more significantly to dust heating. As noted by Calzetti et al. (2010), the correlation between PAH emission and metallicity leads to another source of increased uncertainty in the link between star formation and IR dust emission in evolved sources.

Zoom In Zoom Out Reset image size

Total IR luminosity is also closely related to these SF diagnostics, and has a very similar slope and scatter as L_PAH, for all three parameters and for both the absolute and mass-weighted relations. For the purpose of modeling high-z and photometric SEDs, it is useful to understand how the PAH-to-total IR ratio changes with these properties. Figure 15 (lower panels) shows L_PAH/L_TIR versus these same SF diagnostics, with dashed and dotted lines showing the medians for the SF and AGN/composite subsamples, respectively. In all SF diagnostics there is a significant but highly scattered trend, with more evolved galaxies and AGNs showing weaker PAH features for a given L_TIR. It is not possible to extricate the effects of age and AGN status on L_PAH/L_TIR from these data, although as shown in OD09 both seem to play a part. More specifically, short wavelength PAH features (6.2 μm and 7.7 μm) are diminished in more evolved galaxies and AGNs, while longer wavelength features are not. This is seen in the PAH ratios (OD09) and with respect to L_TIR (T10). The trend is between L_PAH and L_TIR and star formation is still weakly significant for SF galaxies alone, with AGNs excluded, although this is entirely driven by a small number of heavily star-forming sources with very high L_PAH/L_TIR.

The SSGSS IR spectra, with its coverage of all major PAH features, combined with MIPS 70 μm and 160 μm photometry, enables us to fit detailed dust models to determine dust mass. We use the models of Draine & Li (2007), which calculate IR emission spectra for starlight-heated dust consisting of amorphous silicate and graphite grains, and of PAH grains. SSGSS spectra and photometry are fit to template libraries derived from these models (B. T. Draine & A. Li 2010, private communication) to estimate the distribution of illuminating starlight intensity and the relative abundance of PAH. These are combined with MIPS fluxes and the relations detailed in Draine & Li (2007) to determine the total dust mass.

Based on these models, the median dust mass for the sample is M_dust = 8.5 × 10⁷ M_☉, with 90% of the sample in the range 10⁷–10⁹ M_☉.

Figure 16 (left panel) shows the calculated dust masses for the SSGSS sample versus SFR_e, with color coding to indicate AGN type by BPT designation. Also plotted are the results of da Cunha et al. (2010; D10), who fit the GALEX–SDSS–2MASS–IRAS photometry of SDSS and SINGS galaxies with SEDs derived from population synthesis and dust attenuation models (da Cunha et al. 2008) to calculate both dust mass and SFR. SSGSS dust masses are very similar to the D10 SDSS sample. The SINGS dust masses are roughly consistent with the range of dust masses calculated using the Draine & Li (2007) models for a subsample of SINGS galaxies (Draine et al. (2007), which reinforces the comparability of the two models.

Zoom In Zoom Out Reset image size

M_dust is highly correlated with SFR_e, with an almost linear relationship that is similar to that found by D10. The best-fit regression line is

$$\begin{eqnarray} &&M_{{\rm dust}}(M_{\odot }) = (2.0 \pm 0.3) \times 10^7 {\rm SFR}_e(M_{\odot }\,{\rm yr^{-1}})^{(0.95 \pm 0.11)}.\nonumber\\ \end{eqnarray} \tag{ 2 }$$

We note that D10 find a tighter relation between these two parameters for SDSS galaxies. This may indicate a more accurate SFR estimation by D10 over the optical emission line estimate, due to their use of the full FUV-to-FIR SED.

Figure 16 (middle panel) shows dust-to-stellar mass fraction versus SSFR for SSGSS, again including the results of D10. There is a significant positive correlation between dust fraction and SSFR (τ = 0.24; correlation significance: p = 0.001), although here there is significantly more scatter than observed by D10 in their sample. The best-fit regression line is

$$\begin{equation} M_{{\rm dust}}/M_{\odot } = (2.5 \pm 1.0) \times 10^5 {\rm SSFR}({\rm yr^{-1}})^{(0.8 \pm 0.2)}. \end{equation} \tag{ 3 }$$

Figure 16 (right panel) shows the ratio of dust mass to SFR, which, given the tight correlation between SFR and gas mass, may be considered a proxy for the dust-to-gas mass ratio, and so traces the enrichment of the interstellar medium. The SSGSS results follow the general trend of the D10 sample, (τ = −0.175; correlation significance: p = 0.015); however, the careful sample selection of SSGSS means that it contains fewer of the rare, extreme SF galaxies, and the UV selection eliminates the oldest red, dead galaxies; the result is a more restricted range of SSFR that makes it difficult to verify the slope of this relation.

These strong correlations between galaxys' dust content and star formation rate—and especially between their absolute values—supports the idea that SFR provides a useful proxy for dust mass, as found by D10. The careful sample selection of SSGSS adds weight to this result.

As noted by D10, these relations can be interpreted in evolutionary terms. The ISM is enriched by dust with ongoing star formation, and at the same time gas mass decreases, resulting in a drop in SFR. With this drop, the production of dust is quickly superseded by destruction of grains in the ISM, resulting in the positive correlation between dust fraction and SSFR (Figure 16, middle). At the same time, this reduction in dust content with galaxy age appears to be outpaced by the reduction in gas content and the associated drop in SSFR, resulting in the anticorrelation between dust-to-gas mass ratio versus SSFR (Figure 16, right panel).

In this scenario, the balance between ISM enrichment and the destruction of dust by shocks and by the ambient radiation field defines the direction and slope of these relations. The BPT emission line ratios, [O iii]/Hβ and [N ii]/Hα, are sensitive to the hardness of the ambient radiation field and the level of ISM enrichment, respectively, and so may shed light on the nature of this balance. Additionally, as seen in Figure 16, galaxies with BPT-designated AGN components have lower dust fractions than SF galaxies. Although this may be purely due to the increased incidence (and observability) of AGNs in older galaxies, the difference in dust fraction with SSFR seems to be divided surprisingly cleanly down AGN lines. It is possible that AGN shocks and hard UV photons are important in destroying small dust grains, given their potential affect on PAH grain size distributions (Smith et al. 2007; OD09; T10)

Figure 17 (upper panels) shows dust-to-stellar mass fraction versus [O iii]/Hβ and [N ii]/Hα, as well as the distribution of dust fractions by BPT designation. Both ratios show a scattered but significant inverse correlation, and in both cases there is no significant trend taking either SF galaxies or AGNs separately.

Zoom In Zoom Out Reset image size

[O iii]/Hβ provides a good proxy for the hardness of the UV radiation field, and so it is tempting to link dust grain destruction with an increase in hard UV photons, although it is not possible to extricate this effect from the drop in dust production with diminishing star formation. Notably, SF galaxies with hard UV fields ([O iii]/Hβ ≳ 0.5) all have significantly higher dust fractions than AGNs with similar [O iii]/Hβ. These same galaxies have among the highest SSFRs, implying that either the radiation field has little to do with grain destruction, or that dust production due to heavy star formation in these galaxies more than compensates for such depletion.

[N ii]/Hα is sensitive to gas-phase metallicity, and so may be expected to have a positive correlation with dust-to-stellar mass fraction. The observed negative correlation supports the idea that the drop in dust content of these galaxies (M_dust/M_*) outpaces the increase in the gas-phase metallicity (and corresponding dust-to-gas mass fraction) as these galaxies evolve.

Figure 17 (lower panels) shows the ratio of dust mass to SFR versus BPT line ratios. There is no correlation between this and any of the line ratios, and no trend with AGN status. This is expected; while dust-to-stellar mass fraction decreases with stellar population age, gas content and the associated SFR probably decrease more quickly, and so the ratio of dust mass to SFR may not drop.

Dust masses for the entire SSGSS sample are available as part of the data release (see Section 8.2)

SSGSS is a Spitzer Legacy Program, and all spectroscopy is available for download from the Spitzer Data Archive. The archive is accessed via Spitzer's Leopard software. The Web interface for the archive and link to the Leopard download page can be found at http://archive.spitzer.caltech.edu/.

The data release includes low-resolution spectra of the 100 galaxies in the full SSGSS sample and high-resolution spectra of the 33 galaxies in the bright SSGSS sample. These are reduced as described in Section 5. We include spectra with slit positions both combined and uncombined, and with orders and (for low-res) SL and LL modules both stitched and unstitched.

In addition to the spectroscopic data, a range of complementary data products are available via the SSGSS Web site (http://astro.columbia.edu/ssgss).

UV through IR imaging data: including FUV and NUV from GALEX; ugriz from SDSS; IRAC 3.6, 4.5, 5.8, and 7.8 μm bands; MIPS 24, 70, and 160 μm bands; and 16 μm IRS peak-up from SSGSS, with corresponding uncertainty and/or background images.

Matched catalogs of SDSS properties: these are from the SDSS studies at the MPA/JHU group⁹ (Brinchmann et al. 2004), and are drawn from data release 4 of their catalogs. These include measured and derived parameters such as magnitudes, line strengths, Petrosian radii, SFRs, and index strengths, as well as magnitudes K-corrected using the method of Blanton et al. (2006), stellar masses, mass-to-light ratios, dust attenuations, etc. from Kauffmann et al. (2003b), and gas-phase metallicities from Tremonti et al. (2004).

PAH strengths: integrated intensities for the most prominent PAH features and complexes at 6.2 μm, 7.7 μm, 8.6 μm, 11.3 μm, and 17 μm, as well as total PAH intensity, measured by the PAHFIT (Smith et al. 2007) decompositions of SSGSS spectra as described in OD09.

Emission line strengths: line strengths from both the low-res and hi-res spectra from PAHFIT decompositions.

Derived properties: dust masses (see Section 7.4), silicate strength (see Section 7.2), and silicate optical depth from PAHFIT decompositions.