A New Calibration of Star Formation Rate in Galaxies Based on Polycyclic Aromatic Hydrocarbon Emission
DDraft version September 4, 2019
Typeset using L A TEX default style in AASTeX62
A New Calibration of Star Formation Rate in Galaxies Based onPolycyclic Aromatic Hydrocarbon Emission
Yanxia Xie and Luis C. Ho
2, 3 Kavli Institute for Astronomy and Astrophysics, Peking University, Beijing 100871, China [email protected] Kavli Institute for Astronomy and Astrophysics, Peking University, Beijing 100871, China [email protected] Department of Astronomy, School of Physics, Peking University, Beijing 100871, China
ABSTRACTPolycyclic aromatic hydrocarbon (PAH) emission has long been proposed to be a potential star for-mation rate indicator, as it arises from the photodissociation region bordering the Str¨omgren sphere ofyoung, massive stars. We apply a recently developed technique of mid-infrared spectral decompositionto obtain a uniform set of PAH measurements from
Spitzer low-resolution spectra of a large sample ofstar-forming galaxies spanning a wide range in stellar mass ( M (cid:63) ≈ − . M (cid:12) ) and star formationrate ( ∼ . − M (cid:12) yr − ). High-resolution spectra are also analyzed to measure [Ne II] 12.8 µ m and[Ne III] 15.6 µ m, which effectively trace the Lyman continuum. We present a new relation betweenPAH luminosity and star formation rate based on the [Ne II] and [Ne III] lines. Calibrations are givenfor the integrated 5–15 µ m PAH emission, the individual features at 6.2, 7.7, 8.6, and 11.3 µ m, as wellas several mid-infrared bandpasses sensitive to PAH. We confirm that PAH emission is suppressed inlow-mass, dwarf galaxies, and we discuss the possible physical origin of this effect. Keywords: galaxies: formation — galaxies: starburst — galaxies: dwarf — (ISM:) dust, extinction INTRODUCTIONThe star formation rate (SFR) is vital to understand how galaxies acquire their mass, but it is a challenging quantityto measure in external galaxies, wherein it can only be inferred through indirect estimators. Dust-corrected ultraviolet(UV) continuum emission and optical line emission ([O II] λ α ) have been used frequently to probe star formationon timescales of a few to ∼
200 Myr (e.g., Gallagher et al. 1989; Kennicutt 1998; Hao et al. 2011; Kennicutt & Evans2012). These UV and optical tracers, however, suffer sensitively from dust extinction and are unavailable for highlyobscured systems, such as ultraluminous infrared (IR) galaxies (ULIRGs). Moreover, the interstellar environments ofsome systems (e.g., dwarf galaxies and outer disks of spirals) may experience severe Lyman continuum photon leakage,posing a major complication for SFR indicators based on recombination lines (Hunter et al. 2010; Rela˜no et al. 2012;Calzetti 2013).Dust absorbs the ionizing continuum from massive stars and reradiates the energy in the IR. Thus, barring situationswhere IR cirrus emission heated by old stars dominates (Kennicutt et al. 2009), or in situations where dust opacitydrops toward young stars (Calzetti 2001), the total IR luminosity ( L IR ) effectively probes the SFR on timescales of ∼ yr. The full IR spectral energy distribution is not always accessible for most galaxies, and, in practice, themonochromatic luminosity in a single band is often adopted as a SFR indicator (e.g., 24 µ m: Alonso-Herrero et al.2006, P´erez-Gonz´alez et al. 2006, Rela˜no et al. 2007, Rieke et al. 2009; 70 and 160 µ m: Bavouzet et al. 2008, Calzettiet al. 2010). Furthermore, the IR window contains several prominent fine-structure lines that have been proposed asSFR estimators. As discussed by Ho & Keto (2007; see also Zhuang et al. 2019), the fine-structure lines [Ne II] 12.8 µ mand [Ne III] 15.6 µ m are especially promising. [C II] 158 µ m, a major coolant associated with photodissociation regions(PDRs), has long been recognized as an important tracer of recent star formation (e.g., Stacey et al. 1991, 2010; DeLooze et al. 2011; Vallini et al. 2015), its importance highlighted all the more with the advent of the Atacama LargeMillimeter/sub-millimeter Array, which can detect [C II] out to very high redshifts (e.g., Pentericci et al. 2016; Bradaˇcet al. 2017).Similar to [C II] 158 µ m, the mid-IR emission features from polycyclic aromatic hydrocarbons (PAHs) also arisefrom PDRs and hence potentially can serve as a SFR indicator, as attested by the empirical correlation between PAH a r X i v : . [ a s t r o - ph . GA ] S e p emission and other star formation tracers (e.g., F¨orster Schreiber et al. 2004; Peeters et al. 2004). The PAH bandsare rich, pervasive, and energetically significant, dominating the mid-IR spectrum from ∼ µ m and accountingfor ∼
10% of the IR energy budget in starburst galaxies (e.g., Smith et al. 2007; Shipley et al. 2013). These attributesrender PAH emission a powerful tool for probing star formation over a wide range of redshifts. In view of theseadvantages, there have many previous efforts to devise a quantitative calibration of the SFR in star-forming galaxiesbased on the strength of PAH emission (e.g., Wu et al. 2005; Farrah et al. 2007; Pope et al. 2008; Treyer et al. 2010;Diamond-Stanic & Rieke 2012; Shipley et al. 2016).This study offers a new PAH-based SFR calibration, using a uniform set of PAH measurements derived from atechnique recently introduced by Xie et al. (2018). Our technique utilizes a theoretical PAH template calculatedfrom Draine & Li (2007) to decouple the complex PAH features from other dust components in the mid-IR spectralregion covered by the
Spitzer
Infrared Spectrograph (IRS; Houck et al. 2004; Werner et al. 2004), allowing us toobtain robust measurements of the integrated PAH emission, including the associated PAH continuum. We presentcalibrations for the total (5–15 µ m) PAH luminosity as well as for the luminosity of certain key individual PAH bands.Our calibration is anchored by the relationship between SFR and the strength of the mid-IR lines of [Ne II] 12.8 µ mand [Ne III] 15.6 µ m, as originally proposed by Ho & Keto (2007) and recently revisited by Zhuang et al. (2019). Thecommon bandpass of PAH and neon emission minimizes complications due to dust obscuration and aperture mismatch.We show that this approach produces SFRs that are generally consistent with those derived from the IR continuum,extinction-corrected H α , or from the combination of H α + 24 µ m. Special care must be taken, however, to treat dwarfgalaxies, which tend to show a deficit in PAH.We introduce the data in Section 2. The measurement of the PAH and neon lines are given in Section 3. Section4 presents the new SFR calibration, which is discussed in Section 5. Readers who are only interested in the finalcalibration can go directly to Section 4. A summary is given in Section 6. Throughout this paper, we adopt thecosmological parameters Ω m = 0 . Λ = 0 . H = 67 . − Mpc − for a ΛCDM cosmology (PlanckCollaboration et al. 2016). For galaxies with z < .
02, the luminosity distance is calculated from the heliocentricvelocity corrected for the flow-field model of Mould et al. (2000). CALIBRATION SAMPLE2.1.
Galaxy Selection
Our calibration data are gathered from the literature and from the
Spitzer archive. The sample comprises local dwarfgalaxies, normal star-forming galaxies, and starburst galaxies that span a large dynamical range in luminosity, stellarmass, and SFR. We require the selected galaxies to have high-quality IRS low-resolution and high-resolution spectra,the former providing the broad spectral coverage necessary to cover the widespread PAH emission and the latter havingsufficient resolution to properly deblend [Ne II] 12.8 µ m and [Ne III] 15.6 µ m from neighboring contaminating features.After extensive experimentation with the IRS database, we determined that reliable measurements can be obtainedfrom spectra that have a minimum signal-to-noise ratio per pixel of ∼ . < z < .
21) star-forming galaxies spanning 2 orders of magnitude in stellar mass, color, and dustattenuation (O’Dowd et al. 2009, 2011). SSGSS contains short-low and long-low IRS spectra, but only a subsetof the 21 brightest galaxies has high-quality short-high spectra.2. The Spitzer SDSS Statistical Spectroscopic Survey (S5), an expanded version of SSGSS, offers 175 additional star-forming galaxies at 0 . < z < . α fluxes greater than 3 × − erg s − cm − . CombiningSSGSS and S5 results in 196 normal star-forming galaxies, which cover total IR luminosities L IR ≈ − L (cid:12) and stellar masses M (cid:63) ≈ − M (cid:12) .3. To incorporate more luminous, more massive objects, we add IR-luminous galaxies from Farrah et al. (2007),who presented 53 ULIRGs observed with IRS. As we are concerned with objects powered by star formation, we FR Traced by PAHs -3 -2 -1 Redshift6789101112 l og ( M ∗ / M O • ) SFGs BCDs ULIRGs
Figure 1.
Distribution of redshift and stellar mass of the sample, which comprises star-forming galaxies (SFGs), blue compactdwarfs (BCDs), and ultraluminous IR galaxies (ULIRGs). only retain the 13 objects that show no obvious signs of strong active galactic nuclei based on diagnostics in theX-rays (Iwasawa et al. 2011; Torres-Alb`a et al. 2018), optical (Veilleux et al. 1995, 1999), and IR (Armus et al.2007; Farrah et al. 2007). Among them, three have stellar mass measurements in Shangguan et al. (2018); for theremaining 10, we estimate stellar masses from the 2MASS J -band photometry (see Appendix A) following themethodology of Shangguan et al. (2018). The ULIRGs span L IR ≈ . − . L (cid:12) and M (cid:63) ≈ . − . M (cid:12) .4. At the opposite extreme, we include less massive, low-luminosity, low-metallicity dwarf galaxies. A systematicsearch of the Spitzer /IRS archive uncovered 18 dwarf galaxies with suitable observations, 17 of which we retainbecause either PAH or neon is detected. They are all considered blue compact dwarfs (BCDs), including twoof the most metal-poor galaxies known—I Zw 18 with metallicity Z = 1 / Z (cid:12) (Searle & Sargent 1972) andSBS 0335 −
052 with Z = 1 / Z (cid:12) (Izotov et al. 1997). The MPE-JHU catalog gives stellar masses for four ofthe objects, and for the remaining 14 we calculate stellar masses according to the prescription of Bell et al.(2003), using their J -band luminosity and (cid:104) g − i (cid:105) color (Appendix A). The stellar masses of the BCDs rangemostly between M (cid:63) ≈ and 10 M (cid:12) , except for Haro 11 and CG 0752, which are distinctly more massive( M (cid:63) ≈ M (cid:12) ). Haro 11 qualifies as a luminous IR galaxy (Wu et al. 2006), but it is reported to have very lowmetallicity ( Z = 1 / Z (cid:12) ; Bergvall et al. 2000).Figure 1 summarizes the redshift and stellar mass distribution of the combined sample of 226 galaxies (196 normalstar-forming galaxies, 13 ULIRGs, and 17 BCDs; Table 1). The entire sample covers z < . M (cid:63) ≈ − . M (cid:12) . 2.2. Data
The IRS has both a low-resolution and a high-resolution mode, each with short and long slits to cover a differentwavelength range. The short-low mode has a slit size of 3 . (cid:48)(cid:48) × (cid:48)(cid:48) and 3 . (cid:48)(cid:48) × (cid:48)(cid:48) , covering, respectively, 7 . − . µ mand 5 . − . µ m, while the long-low mode covers 19 . − . µ m with the 10 . (cid:48)(cid:48) × (cid:48)(cid:48) slit and 14 . − . µ m withthe 10 . (cid:48)(cid:48) × (cid:48)(cid:48) slit. The resolution varies from λ/ ∆ λ ≈
64 to 128 in each segment. The short-high mode samples9 . − . µ m with the 4 . (cid:48)(cid:48) × . (cid:48)(cid:48) slit, the long-high mode 18 . − . µ m with the 11 . (cid:48)(cid:48) × . (cid:48)(cid:48) slit, both with λ/ ∆ λ ≈ Spitzer /IRS Sources (CASSIS; Lebouteilleret al. 2011, 2015), which adopts a spectral extraction scheme according to the spatial extent of each source. Sourcesnot included in CASSIS were retrieved from the Spitzer Legacy Program Database .Aperture mismatch is a major concern given the relative proximity of our sources and the variety of slit widthsemployed by IRS. We adopt the following strategy to mitigate this complication. Recalling that our primary goal isto calibrate PAH emission relative to the independent SFR indicator based on [Ne II] 12.8 µ m and [Ne III] 15.6 µ m,we note that the bulk of the PAH emission is confined to the ∼ − µ m region covered by the short-low spectrum(see Figure 4a), while both neon lines are contained in the short-high spectrum. Fortunately, the apertures of bothmodes are quite similar, 3 . (cid:48)(cid:48) for short-low and 4 . (cid:48)(cid:48) for short-high, and for the purposes of this study, we will assumethat the difference is unimportant. For the normal star-forming galaxy (SSGSS and S5) sample, aperture correctionhas already been performed for the high- and low-resolution spectra (e.g., O’Dowd et al. 2011). As for the rest, theULIRGs with z (cid:38) .
08 are sufficiently compact to be fully captured by the short-low aperture. Four ULIRGs with z < .
08 and most of the BCDs (median z = 0 . ∼ µ m. We scale thelarger aperture (10 . (cid:48)(cid:48) ) long-low spectrum to match the short-low spectrum using their common overlapping region,making the explicit assumption that the mid-IR spectrum does not change dramatically between these two apertures.This generally holds for nearby galaxies that have been studied with spatially resolved mid-IR observations (e.g.,Maragkoudakis et al. 2018). MEASUREMENTS3.1.
Neon Lines
The [Ne II] 12.8 µ m and [Ne III] 15.6 µ m lines, being narrow and well-separated from nearby features in the short-high spectrum, can be measured straightforwardly using a single Gaussian fit on top of a local continuum. Weselect continuum regions immediately adjacent to either side of each line and interpolate with a linear continuum.The distribution of repeated observations gives the flux and its uncertainty. To quantify the significance of the linemeasurement, we define the quantity S Ne as the ratio of the line flux to the full width at half maximum (FWHM) ofthe line. We consider a line detected if S Ne ≥ σ , where σ is the standard deviation of the local continuum; otherwise,the line is a non-detection, and the 3 σ upper limit is given by fixing the line width to the median value of the detectedsources (FWHM = 1375 km s − for [Ne II] and 1133 km s − for [Ne III]). In total, there are 7 upper limits for [Ne II]and 100 upper limits for [Ne III]. The total flux of neon is the sum of [Ne II] and [Ne III], and neon is regarded asdetected when either line is detected. Errors are estimated from boostrap sampling. Two sources are undetected inneither [Ne II] nor [Ne III]. The extinction-corrected neon luminosities are given in Table 1.3.2. PAH Features
We measure the PAH strength using the methodology of Xie et al. (2018), which is summarized briefly here.After first removing ionic emission lines that are not blended with the main PAH features, we fit the ∼ − µ mspectrum with a four-component model consisting of a theoretical PAH template plus dust continuum representedby three modified blackbodies of different temperatures, all subject to dust attenuation by foreground extinction.Our theoretical PAH spectrum is calculated by adopting a starlight intensity U = 1 times the interstellar radiationfield of the solar neighborhood interstellar medium (Mathis et al. 1983) and grain sizes 3 . < a <
20 ˚A to accountfor stochastic heating of small grains (Draine & Li 2001). The PAH spectrum is relatively insensitive to radiationintensities up to U ≈ . The global best fit is determined using the Levenberg-Marquardt χ -minimization algorithm MPFIT (Markwardt 2009). To estimate the uncertainties of the final PAH spectra, we randomly sample the observedspectra 100 times and repeat the fits; the median and standard deviation of the 100 realizations give the final PAHflux and its 1 σ error. Upper limits (which, in this sample, apply only to some of the BCDs) on the PAH flux are set as3 σ . We further consider the systematic uncertainties associated with the continuum model, which are estimated fromMonte Carlo simulations as described in Xie et al. (2018; their Section 2 and Figure 2). The final error budget on thePAH flux is the quadrature sum of these two contributions. https://irsa.ipac.caltech.edu/data/SPITZER/S5/, https://irsa.ipac.caltech.edu/data/SPITZER/SSGSS/ FR Traced by PAHs − µ m PAH flux, as well as the flux of the commonly used individual PAH bands at6.2, 7.7, 8.6 and 11.3 µ m, adopting a Drude profile and the parameters given in Table 1 of Draine & Li (2007). Toobtain uncertainties for each feature, we repeat the fitting 100 times on randomly simulated PAH spectra and take themedian and standard deviation as the final flux and associated error. Dust extinction can be significant, especially forULIRGs, where silicate absorption at 9.7 µ m can affect the 8.6 and 11.3 µ m PAH features by more than a factor of 10.Mid-IR extinction is generally negligible ( < T a b l e . P h y s i c a l P r o p e r t i e s o f t h e S a m p l e O b j ec t z D L l og M ∗ Z l og τ . l og L [ N e II ] + [ N e III ] [ N e III ] / [ N e II ] f + f + l og S F R N o t e s ( M p c )( M (cid:12) ) + l og ( O / H )( e r g s − )( M (cid:12) y r − ) ( )( )( )( )( )( )( )( )( )( )( )( ) C G ∗ . . + . − . . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . B C D H a r o11 ∗ . . + . − . . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . B C D II Z w ∗ . . + . − . . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . B C D I Z w . . + . − . . + . − . − . + . − . . + . − . > . . . − . + . − . B C D K U G + ∗ . . + . − . . + . − . − . + . − . . + . − . > . . . − . + . − . B C D M r k . . + . − . . + . − . < − . . + . − . . + . − . . + . − . . + . − . − . + . − . B C D M r k ∗ . . + . − . . < − . . + . − . . + . − . . + . − . . + . − . . + . − . B C D M r k ∗ . . + . − . . + . − . < − . . + . − . > . . . − . + . − . B C D M r k ∗ . . + . − . . + . − . < − . . + . − . . + . − . . + . − . . + . − . − . + . − . B C D I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S + ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S + ∗ . . + . − . . + . − . . + . − . . + . − . < . . . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − ∗ . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G M A SS J + . . + . − . . + . − . < − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . . + . − . . + . − . < − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . < . . . . + . − . S F G M A S X J + . . + . − . . + . − . < − . . + . − . < . . . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . < . . . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . . + . − . . + . − . − . + . − . . + . − . . + . − . . + . − . . + . − . . + . − . S F G N o t e — C o l. ( ) O b j ec t n a m e ; a s t e r i s k ( ∗ ) i nd i c a t e s a n o b j ec t w h o s e s t e ll a r m a ss i s c a l c u l a t e d i n t h ec u rr e n t w o r k . C o l. ( ) R e d s h i f t . C o l. ( ) L u m i n o s i t y d i s t a n ce . C o l. ( ) S t e ll a r m a ss . C o l. ( ) M e t a lli c i t y . W ec o ll ec tt h e m e t a lli c i t y m e a s u r e m e n t s o f H a r o11 , N G C , II Z w , U G C , C G , a nd M r k f r o m W u e t a l. ( ) ; M r k , U M , S B S − E , I Z w , M r k , S B S + , S B S + , S B S + , a nd M r k f r o m I z o t o v & T hu a n ( ) ; K U G + f r o m I z o t o v e t a l. ( ) ; S D SS J + f r o m B e r g e t a l. ( ) ; a nd T o l f r o m K o bu l n i c ky & S k ill m a n ( ) . C o l. ( ) D u s t e x t i n c t i o n a t . µ m . C o l. ( ) L u m i n o s i t y o f [ N e II ] + [ N e III ]. C o l. ( ) L u m i n o s i t y r a t i oo f [ N e III ] / [ N e II ]. C o l. ( ) F r a c t i o n a l a bund a n ce o f n e o n i n [ N e II ] s t a t e . C o l. ( ) F r a c t i o n a l a bund a n ce o f n e o n i n [ N e III ] s t a t e . C o l. ( ) S t a r f o r m a t i o n r a t e . C o l. ( ) G a l a xy t y p ec l a ss i fi e db a s e d o nb o l o m e tr i c l u m i n o s i t y : s t a r - f o r m i n gga l a xy ( S F G ) ; b l u ec o m p a c t d w a r f ( B C D ) ; u l tr a l u m i n o u s i n f r a r e d ga l a xy ( U L I R G ) . T h ec o m p l e t e t a b l e i s a v a il a b l e i n t h e o n li n e j o u r n a l. FR Traced by PAHs T a b l e . P AH L u m i n o s i t y i n D i ff e r e n t B a nd s O b j ec t l og L P AH ( . µ m ) l og L P AH ( . µ m ) l og L P AH ( . µ m ) l og L P AH ( . µ m ) l og L P AH ( − µ m ) N o t e s ( e r g s − )( e r g s − )( e r g s − )( e r g s − )( e r g s − ) ( )( )( )( )( )( )( ) C G . + . − . . + . − . . + . − . . + . − . . + . − . B C D H a r o1141 . + . − . . + . − . . + . − . . + . − . . + . − . B C D II Z w . + . − . . + . − . < . . + . − . . + . − . B C D I Z w < . < . < . < . < . B C D K U G + < . < . < . < . < . B C D M r k < . < . < . < . < . B C D M r k . + . − . . + . − . . + . − . . + . − . . + . − . B C D M r k . + . − . . + . − . < . . + . − . . + . − . B C D M r k < . < . < . . + . − . < . B C D N G C . + . − . . + . − . . + . − . . + . − . . + . − . B C D I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S + . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S + . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G I R A S − . + . − . . + . − . . + . − . . + . − . . + . − . U L I R G M A SS J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G M A S X J + . + . − . . + . − . . + . − . . + . − . . + . − . S F G N o t e — C o l. ( ) O b j ec t n a m e . C o l. ( ) L u m i n o s i t y o f . µ m P AH f e a t u r e . C o l. ( ) L u m i n o s i t y o f . µ m P AH f e a t u r e . C o l. ( ) L u m i n o s i t y o f . µ m P AH f e a t u r e . C o l. ( ) L u m i n o s i t y o f . µ m P AH f e a t u r e . C o l. ( ) L u m i n o s i t y o f P AH i n − µ m b a nd . C o l. ( ) G a l a xy t y p ec l a ss i fi e db a s e d o nb o l o m e tr i c l u m i n o s i t y : s t a r - f o r m i n gga l a xy ( S F G ) ; b l u ec o m p a c t d w a r f ( B C D ) ; u l tr a l u m i n o u s i n f r a r e d ga l a xy ( U L I R G ) . T h ec o m p l e t e t a b l e i s a v a il a b l e i n t h e o n li n e j o u r n a l. L [Ne II] (erg s -1 )10 L [ N e III] ( e r g s - ) SFGs BCDs ULIRGs (a) L [Ne II]+[Ne III] (erg s -1 )10 L P AH ( e r g s - ) PAH (5-15 µ m) (b) M ∗ / M O • )-1.5-1.0-0.50.00.5 l og L P AH - l og L [ N e II] + [ N e III] (c)
Figure 2. (a) Relationship between [Ne II] 12.8 µ m and [Ne III] 15.6 µ m luminosity; the solid line shows the 1:1 relation.(b) Correlation between the luminosity of [Ne II]+[Ne III] ( L [Ne II]+[Ne III] ) and total PAH emission ( L PAH ) in the wavelengthrange 5 − µ m. The dashed line represents the median value of L [Ne II]+[Ne III] /L PAH for normal star-forming galaxies andstarburst-dominated ULIRGs. (c) The residuals of panel (b), log L PAH − log L [Ne II]+[Ne III] , are plotted as a function of stellarmass. 4. SFR CALIBRATION4.1.
SFR Based on PAH Emission
Our PAH-based SFR estimator is directly calibrated against SFRs for our calibration sample determined from the[Ne II] 12.8 µ m and [Ne III] 15.6 µ m lines, which for our objects span ∼ − ≈ .
3, while for the low-massBCDs the ratio is ∼
20 times higher, at [Ne III]/[Ne II] ≈ . − µ m ( L PAH ) correlates closely with thetotal neon emission ( L [Ne II]+[Ne III] ) when L PAH (cid:38) erg s − . This regime is mostly occupied by massive galaxies.By contrast, BCDs, which in our sample have L PAH (cid:46) erg s − , uniformly exhibit a deficit in PAH emission relativeto neon. As discussed in Section 5.1, the main physical driver for the PAH deficit in BCDs is metallicity, or, indirectly,stellar mass, given the relationship between the two (e.g., Tremonti et al. 2004). The onset of the PAH deficit occursbelow a stellar mass of M ∗ ≈ M (cid:12) , and the magnitude of the deficit decreases systematically with increasing M ∗ ,albeit with significant scatter (Figure 2c).Zhuang et al. (2019) updated the neon-based SFR relation of Ho & Keto (2007) to explicitly include the effect ofmetallicity: SFR ( M (cid:12) yr − ) = 4 . × − ( Z (cid:12) /Z ) (cid:20) L [Ne II]+[Ne III] (erg s − ) f + + 1 . f +2 (cid:21) , (1)where Z is the gas-phase metallicity of the galaxy, and f + and f +2 are the fractional abundances of singly and doublyionized neon. The fractional abundances of ionized neon depend on a number of factors, as described in Zhuang etal. (2019), who performed photoionization calculations over a wide range of metallicities and star formation histories,assuming a Salpeter (1955) stellar initial mass function. Zhuang et al. (2019; their Equation 5) give an explicitfitting function to estimate f + and f +2 based on the observed [Ne III]/[Ne II] ratio. Uncertainties are estimatedfrom bootstrap resampling. The derived fractional abundances populate well the full range of values predicted fromthe photoionization models. For galaxies with [Ne III] undetected, we adopt the values of f + and f +2 in the limit[Ne III]/[Ne II] → −∞ , whereas we adopt the values for [Ne III]/[Ne II] → ∞ when [Ne II] is undetected. For galaxieswith neither line detected, we take the peak values of f + = 0 .
75 and f +2 = 0 .
15 (Ho & Keto 2007).We estimate the metallicities for the massive galaxies from the mass-metallicity relation of Kewley & Ellison (2008),who derived metallicities based on a combination of the N2O2 (Kewley & Dopita 2002) and N2 (Pettini & Pagel 2004)methods. The N2O2 method yields higher absolute metallicities, while the N2 approach gives lower values. We take
FR Traced by PAHs M ∗ < . M (cid:12) (e.g., Jimmy et al. 2015; Yates et al. 2019); thus, for the BCDs, including theirtwo massive members, we use direct metallicity measurements from the literature (Table 1).Figure 3 shows the empirical correlation between our neon-based SFRs and the integrated 5 − µ m PAH luminosity.Massive, star-forming galaxies and ULIRGs trace a tight, linear relation over nearly 4 orders of magnitude in L PAH and3 orders of magnitude in SFR. As anticipated, low-mass BCDs define a distinctly offset sequence with larger scatter.We parameterize the relation between SFR and PAH luminosity aslog SFR ( M (cid:12) yr − ) = α (log L PAH −
43) + β, (2)where L PAH is in units of erg s − , β = β h for M ∗ ≥ M (cid:12) , and β = β l for M ∗ < M (cid:12) (Table 3). We performregression analysis using the IDL procedure
LINMIX ERR (Kelly 2007) to derive the best-fit model parameters andcalculate the scatter of the above correlation. Based on Bayesian inference,
LINMIX ERR accounts for measurementerrors in the linear regression by computing the likehood function for the observed data, and upper limits in thedependent variable (SFR) are properly treated. The best-fit model parameters are each summarized in the posteriordistribution, where a normal density is adopted as priors given the observed dataset. The correlation is almost perfectlylinear for massive galaxies (e.g., α = 0 . ± .
034 for 5–15 µ m PAH).Since ∼
60% of the PAH emission in BCDs are upper limits, we constrain the offset in the SFR − L PAH relationbetween BCDs and high-mass galaxies by fixing the slope of the former to that of the latter. We estimate the offsetby bootstrap resampling using two cases, which likely approximate the maximum and minimum offsets of the twogalaxy samples. Specifically, we resample 500 times the PAH luminosities of the undetected BCDs, from truncatednormal distributions between zero and the 3 σ upper limits, or uniform distributions between zero and 3.5 σ . For thedetections, we resample L PAH from normal distributions with σ set to the measurement errors. We then multiply theresampled L PAH with α and subtract them from the SFR to estimate the distribution of β l ; the median and standarddeviation of the 500 realizations give the final β l and its 1 σ error. The results from the two cases agree within theiruncertainties. The case of the uniform distribution gives β l − β h = 1 . ± .
083 for 5–15 µ m PAH. We tabulate thesolutions from the uniform distribution in Table 3, and show the average offset on Figure 3.In addition to the integrated 5 − µ m PAH emission, Table 3 also gives the best-fit parameters for several commonlyused individual PAH bands at 6.2, 7.7, 8.6, and 11.3 µ m. Our PAH-based SFR estimator is applicable to galaxiesspanning over 5 orders of magnitude in M (cid:63) ( ∼ − . M (cid:12) ) and more than 4 orders of magnitude in SFR( ∼ . − M (cid:12) yr − ), among them dwarf galaxies, main-sequence star-forming galaxies, and vigorous starbursts.The total scatter of the SFR- L PAH relation for the high-mass sample is quite small, with (cid:15) h (cid:46) . (cid:15) l ≈ . L PAH relation is remarkably tight for high-mass galaxies. Six out of the 17 nearby BCDs suffer from a slightaperture mismatch between the short-low and short-high spectra, but removing these six objects from the samplehardly perturbs the final fits. Lastly, we verify that the final calibration is also robust with respect to our treatmentof upper limits. Restricting the calibration only to massive galaxies with both neon lines detected yields a fit verysimilar to that of the full sample that includes upper limits.We end this section with a cautionary note. The SFR calibration given above is based on spatially integrated, globallyaveraged spectra and implicitly assumes that all of the PAH emission is associated exclusively with star-forming regions.Spatially resolved studies of nearby galaxies reveal an imperfect correspondence between PAH-emitting regions andH II regions (Maragkoudakis et al. 2018). PAH emission also arises from inter-arm regions with no obvious associationwith young stars (see, e.g., Figure 8 in Dale et al. 2009; Crocker et al. 2013), as well as in environments where evolvedstellar populations dominate the heating (e.g., Bressan et al. 2006; Kaneda et al. 2008).4.2.
SFR Based on Photometric Bands l og ( M ∗ / M O • ) L PAH (erg s -1 )10 -1 SF R ( M O • y r - ) PAH (5-15 µ m) SFGs BCDs ULIRGs Haro 11 CG 0752
Figure 3.
The dependence of L PAH on SFR. The SFRs are based on L [Ne II]+[Ne III] (Equation 1). The dashed line gives thebest-fit regression for the sample of massive ( M (cid:63) ≥ M (cid:12) ) star-forming galaxies and ULIRGs. The dotted line represents thezero point offset for the low-mass ( M (cid:63) < M (cid:12) ) BCDs. Two massive BCDs (Haro 11 and CG 0752) are highlighted. The datapoints are color-coded according to stellar mass. Table 3.
Best-fit Parameters for the SFR Calibration in PAH BandsBand α β h (cid:15) h β l (cid:15) l (1) (2) (3) (4) (5) (6)PAH 5 − µ m 0.948 ± ± ± ± ± µ m 1.052 ± ± ± ± ± µ m 0.999 ± ± ± ± ± µ m 0.992 ± ± ± ± ± µ m 1.013 ± ± ± ± ± Note —Best-fit parameters for log SFR = α (log L PAH −
43) + β , where β = β h for M ∗ ≥ M (cid:12) and β = β l for M ∗ < M (cid:12) . Col. (1) PAHband used in calibration. Col. (2) Slope of the massive galaxies ( M ∗ ≥ M (cid:12) ). Col. (3) Zero point of the massive galaxies. Col. (4) Total scatterof the massive galaxies. Col. (5) Zero point of the low-mass ( M ∗ < M (cid:12) ) dwarf galaxies. Col. (6) Total scatter of the dwarf galaxies. PAH emission is sufficiently prominent and widespread throughout the mid-IR spectrum that for local galaxies themore conspicuous features can be captured using the photometric bands of existing and upcoming facilities. Figure 4highlights some examples applicable to z ≈ Spitzer
IRAC4 8 µ m, WISE
W3 12 µ m, and JWST
F770W at 7.7 µ mand F1130W at 11.3 µ m), but obviously appropriately matched bandpasses can be considered for higher redshifts.We calibrate the PAH-sensitive photometric bands against SFRs derived from our extinction-corrected L [Ne II]+[Ne III] ,using synthetic photometry obtained by convolving the response curves of the respective filters with the rest-framespectra of our objects. For illustration, Figure 5 shows the case of WISE /W3; the other filters behave qualitativelysimilarly. The relation between SFR and the photometric bands is highly nonlinear, even for the massive galaxiesalone; a linear fit yields a reduced χ ν = 1 .
32 for massive galaxies. A third-order polynomial gives an improved fit
FR Traced by PAHs -2 -1 R e s pon s e C u r v e f ν / f ν ( . µ m ) Rest Wavelength ( µ m) (b) W3 IRAC4 F770W F1130W -1 PAH Features (a)
12 12.7 16.4 17.1
Silicates Silicates
Figure 4. (a) Example mid-IR spectrum of 2MASX J10470453+5620253, illustrating the locations of the prominent PAH andsilicate features. (b) Response curves of filters that are sensitive to the PAH features, including (green)
WISE
W3 (12 µ m)(blue) Spitzer
IRAC4 (8 µ m), and (red) JWST
F770W (7.7 µ m) and F1130W (11.3 µ m). Table 4.
Best-fit Parameters for the SFR Calibration in Photometry BandsBand a b c d (cid:15) (1) (2) (3) (4) (5) (6)IRAC4 1.494 ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± Note —Best-fit parameters for log SFR = a + bx + cx + dx , where x = log L band −
43. Col. (1) Band used in calibration. Col. (2) Coefficient a . Col. (3) Coefficient b . Col. (4) Coefficient c . Col. (5) Coefficient d . Col. (6) Total scatter. with reduced χ ν = 1 .
16. Instead of treating the massive and dwarf galaxies separately, as we did for the SFR − L PAH relation (Figure 3), for simplicity we fit the entire sample with a third-order polynomial,log SFR ( M (cid:12) yr − ) = a + bx + cx + dx , (3)where x = log L band −
43 and L band is in units of erg s − . The total scatter is ∼ . − . DISCUSSION5.1.
The PAH Deficit in Dwarf Galaxies
Most of the BCDs in our sample have conspicuously weak PAH emission, and the degree of PAH deficit is stronglytied to stellar mass. The PAH deficit in Figure 2c is illustrated for the integrated PAH emission from 5–15 µ m, butwe confirm that the effect is qualitatively similar for the PAH 7.7 and 11.3 µ m features (see Table 3 for quantitative2 l og ( M ∗ / M O • ) L W3 (erg s -1 )10 -1 SF R ( M O • y r - ) SFGs BCDs ULIRGs
Figure 5.
The dependence of L W3 on SFR. The SFRs are based on L [Ne II]+[Ne III] (Equation 1). The dashed line gives thebest-fit third-order polynomial. The data points are color-coded according to stellar mass. estimates). A number of previous studies have noted the general tendency for dwarf galaxies to exhibit low PAHequivalent widths (e.g., Peeters et al. 2004; Engelbracht et al. 2005; Madden et al. 2006; Wu et al. 2006; Shipley et al.2016). While the true cause of the reduction in PAH emission is still unknown, the association of weak PAH emissionwith prominent [Ne III]/[Ne II] in low-mass, low-metallicity dwarf galaxies suggests that the harder radiation fieldsof subsolar metallicity environments may not be conducive to the survival of PAH molecules (e.g., Plante & Sauvage2002). Destruction of PAHs by shocks has also been implicated, in view of the elevated levels of [Fe II] 26 µ m emissionreported in some systems (O’Halloran et al. 2006). Perhaps supernova-driven shocks are more prevalent in dwarfgalaxies if their low metallicities lead to a more top-heavy initial mass function (e.g., Bromm et al. 2002). However, acareful search for [Fe II] 26 µ m in our sample of BCDs detected this feature in only five out of 17 objects, at a medianvalue of [Fe II]/[Ne II] = 0.13 that lies significantly below that of previous observations (O’Halloran et al. 2006). Thepoor statistics of our sample preclude us from discussing this issue further.Might the depression of PAH emission be attributed simply to the overall low metal content in dwarf galaxies?Not likely, not unless carbon is preferentially depleted relative to neon. O’Halloran et al. (2006) speculate that dwarfgalaxies may be too young to host a significant population of asymptotic giant branch stars, whose stellar winds arethought to be a main site of dust formation (Morgan & Edmunds 2003). The ultra-metal-poor dwarf galaxy I Zw 18presents a strong counterargument to this hypothesis: its mid-IR spectrum is completely devoid of PAH emission, andyet it is clearly evolved enough to host abundant asymptotic giant branch stars (Izotov & Thuan 2004).We offer an alternative, though related, proposal. We suggest that PAHs are depleted in dwarf galaxies principallybecause low-metallicity environments lack sufficient dust grains to shield these large, fragile molecules from photodisso-ciation by UV radiation. This is analogous to the mechanism commonly invoked to explain the deficit of CO emissionin this class of galaxies (e.g., Leroy et al. 2011; Magdis et al. 2012; Bolatto et al. 2013). Quantitative calculations areneeded to assess the viability of this hypothesis.5.2. Comparison with Previous PAH-Based SFR Calibrations
FR Traced by PAHs
13A number of studies have presented empirical calibrations of SFR based on individual PAH bands at 6.2, 7.7,and 11.3 µ m, or some combinations thereof. For a proper comparison, we only consider previous works that givecalibrations involving the integrated luminosity of the PAH bands, not those based on measurements of peak lineintensity (e.g., Houck et al. 2007; Weedman & Houck 2008; Sargsyan & Weedman 2009), which would be significantlyaffected by the underlying continuum and extinction. Furthermore, we convert all calibrations to the Salpeter initialmass function and infer that they pertain to star formation within 20 Myr, which contributes most of the ionizingflux (Kennicutt 1998). Still, quantitative comparison between our work and those of others is fraught with difficultybecause of fundamental differences in the manner in which the PAH emission was measured. For example, whereas ourapproach (Xie et al. 2018) utilizes global fitting to simultaneously decompose the PAH emission from the underlyingcontinuum components, the PAH measurements of Farrah et al. (2007) and Pope et al. (2008) are based on localcontinuum fits, and the two approaches can produce quite large differences. Moreover, our analysis properly accountsfor the effects of dust extinction, which can be significant in some systems (e.g., ULIRGs), while these previous studiesdid not. Thus, it is virtually impossible to ascertain the significance of the factor of a few difference between ourresults and those of these studies.The work of Shipley et al. (2016) offers a better point of comparison. Shipley et al. used an aperture-matchedsample of star-forming galaxies to calibrate the PAH bands at 6.2, 7.7, and 11.3 µ m, as well as all possible combinationsof these three features, against SFRs based on H α + 24 µ m. Importantly, their analysis measured the PAH strengthusing the code PAHFIT (Smith et al. 2007), which for individual bands compares favorably to the global-fitting methodof Xie et al. (2018). Our SFR calibrations broadly agree with those of Shipley et al. (2016), both in slope and scatter,although our derived SFRs are a factor of 4, 2, and 1.3 higher for PAH 6.2, 7.7, and 11.3 µ m, respectively. Thedifferences remain even after removing the ULIRGs from the fit. Considering that our calibration sample is more thantwice as large as that of Shipley et al., and we adopt different methods for dust extinction correction, we consider thislevel of discrepancy acceptable.Cluver et al. (2017) calibrated the WISE
W3 band to SFRs determined from total IR luminosity. Our W3-basedSFRs agree with those of Cluver et al. to within 0.15 dex for normal star-forming galaxies and to ∼ . . Cluver et al. selected their ULIRGs based on WISE mid-IR colors, which may still contain some degree ofcontamination by active galactic nuclei (see their Section 3.1). They also applied a procedure to correct W3 for stellaremission. Both factors may contribute to the differences with our results. Brown et al. (2017) published a similarcalibration for
WISE /W3 and S pitzer/IRAC4 using SFRs tied to extinction-corrected H α of 66 local star-forminggalaxies. For massive star-forming galaxies, their calibration generally agrees with ours within a factor of 2 for W3 anda factor of 4 for IRAC4, but their prescription for dwarf galaxies is ∼ SUMMARYThe prominent, pervasive emission features of PAH have the potential to serve as an effective tracer of star formationin galaxies. We apply a recently developed mid-IR spectral decomposition technique to quantify the strength of thePAH features using low-resolution
Spitzer /IRS spectra of a diverse sample of 226 star-forming galaxies, ranging fromlow-mass dwarfs to typical star-forming galaxies and extreme starbursts in ultraluminous IR galaxies. Together withhigh-resolution spectra that yield measurements of the [Ne II] 12.8 µ m and [Ne III] 15.6 µ m lines, which effectivelytrace the Lyman continuum of massive stars, we provide a new PAH-based SFR estimator for galaxies. We presentcalibrations for the integrated 5–15 µ m PAH emission, the individual features at 6.2, 7.7, 8.6, and 11.3 µ m, and, forcompleteness, several mid-infrared bandpasses sensitive to PAH.Our principal results are as follows:1. The extinction-corrected PAH luminosity correlates tightly ( ∼ . M ∗ ≈ − . M (cid:12) ) and star formation activity(SFR ≈ − M (cid:12) yr − ).2. Low-mass ( M ∗ (cid:46) M (cid:12) ) dwarf galaxies exhibit notably weaker PAH emission relative to the neon lines, aswell as larger scatter in their mutual correlation. Our new SFR estimator can be applied to galaxies in this The sample of Cluver et al. (2017) has too few dwarf galaxies to permit a meaningful comparison. ≈ . M (cid:12) yr − . We speculate that the PAH deficit in dwarf galaxies originatesfrom photodissociation of PAH molecules in metal-poor, dust-poor environments.3. Our SFRs are broadly consistent with previous PAH SFRs calibrated against total IR luminosity, extinction-corrected H α , or H α + 24 µ m luminosity, but our calibration sample is extended to include more massive, moreluminous starbursts. We also calibrate PAH-sensitive photometric bands against SFRs, for current ( Spitzer , WISE ) and future (
JWST ) mid-IR facilities.We thank an anonymous referee for helpful comments and suggestions. L.C.H. was supported by the National ScienceFoundation of China (11721303) and the National Key Program for Science and Technology Research and Development(2016YFA0400702). Y.X. is supported by the China Postdoctoral Science Foundation Grant (2016M591007) and theNational Natural Science Foundation of China for Youth Scientist Project (11803001). Y.X. thanks Li Shao for helpin compiling stellar masses from the MPA-JHU catalog; Jinyi Shangguan for providing stellar masses of the BCDsand ULIRGs; Ruancun Li for help with 2MASS photometry for some of the objects; Hassen Yesuf for assistance withstatistical analysis; and Sandra Faber, D. Farrah, Robert Kennicutt, Ming-Yang Zhuang, and Lulu Zhang for insightfuldiscussions. This publication used data products from the Two Micron All-Sky Survey, which is a joint project ofthe University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology,funded by the National Aeronautics and Space Administration and the National Science Foundation. The CornellAtlas of
Spitzer /IRS Sources (CASSIS) is a product of the Infrared Science Center at Cornell University, supportedby NASA and JPL. APPENDIX A. STELLAR MASSES OF ULIRGS AND BCDSThe stellar masses of ULIRGs are derived aslog( M ∗ /M (cid:12) ) = − . M J − M J, (cid:12) ) − .
75 + 0 .
34 ( B − I ) , (A1)where M J and M J, (cid:12) = 3 .
65 mag (Blanton & Roweis 2007) are, respectively, the rest-frame J -band absolute magnitudesof the galaxy and the Sun. We take B − I = 2 . . − . M (cid:12) . For BCDs, the stellar masses are calculated aslog( M ∗ /M (cid:12) ) = − . M J − M J, (cid:12) ) − .
153 + 0 . (cid:104) g − i (cid:105) , (A2)where (cid:104) g − i (cid:105) = 0.738 mag, the color of the low-surface brightness component as given in Meyer et al. (2014). Thetypical mass uncertainty derived in this manner is 0.2 dex (Conroy 2013; Zhang et al. 2016).Four BCDs are not archived in either the 2MASS extended-source catalog or the point-source catalog because oftheir low surface brightness . We measure the J − band photometry of these sources using our own procedure forgalaxy photometry (R. Li et al. , in preparation). All of them have J − band signal-noise-ratio <
5, below the criterionto be included in published catalogs. SBS 1415+437 is detected in the J band at the 2.5 σ level; we can only obtainan upper limit for PAH, but both [Ne II] and [Ne III] are significant on detected. See https://old.ipac.caltech.edu/2mass/releases/allsky/ for more details.
FR Traced by PAHs
Alonso-Herrero, A., Rieke, G. H., Rieke, M. J., et al. 2006,ApJ, 650, 835Arribas, S., Bushouse, H., Lucas, R. A., Colina, L., &Borne, K. D. 2004, AJ, 127, 2522Armus, L., Charmandaris, V., Bernard-Salas, J., et al.2007, ApJ, 656, 148Bavouzet, N., Dole, H., Le Floc’h, E., et al. 2008, A&A,479, 83Bell, E. F., McIntosh, D. H., Katz, N., & Weinberg, M. D.2003, ApJS, 149, 289Bergvall, N., Masegosa, J., ¨Ostlin, G., & Cernicharo, J.2000, A&A, 359, 41Blanton, M. R., & Roweis, S. 2007, AJ, 133, 734Bolatto, A. D., Wolfire, M., & Leroy, A. K. 2013, ARA&A,51, 207Bradaˇc, M., Garcia-Appadoo, D., Huang, K.-H., et al. 2017,ApJL, 836, L2Bressan, A., Panuzzo, P., Buson, L., et al. 2006, ApJL, 639,L55Bromm, V., Coppi, P. S., & Larson, R. B. 2002, ApJ, 564,23Brown, M. J. I., Moustakas, J., Kennicutt, R. C., et al.2017, ApJ, 847, 136Calzetti, D. 2001, PASP, 113, 1449Calzetti, D. 2013, Secular Evolution of Galaxies, 419Calzetti, D., Wu, S.-Y., Hong, S., et al. 2010, ApJ, 714,1256Cluver, M. E., Jarrett, T. H., Dale, D. A., et al. 2017, ApJ,850, 68Conroy, C. 2013, ARA&A, 51, 393Crocker, A. F., Calzetti, D., Thilker, D. A., et al. 2013,ApJ, 762, 79Dale, D. A., Cohen, S. A., Johnson, L. C., et al. 2009, ApJ,703, 517De Looze, I., Baes, M., Bendo, G. J., Cortese, L., & Fritz,J. 2011, MNRAS, 416, 2712Diamond-Stanic, A. M., & Rieke, G. H. 2012, ApJ, 746, 168Draine, B. T., & Li, A. 2001, ApJ, 551, 807Draine, B. T., & Li, A. 2007, ApJ, 657, 810Engelbracht, C. W., Gordon, K. D., Rieke, G. H., et al.2005, ApJL, 628, L29Farrah, D., Bernard-Salas, J., Spoon, H. W. W., et al. 2007,ApJ, 667, 149Ferland, G. J., Chatzikos, M., Guzm´an, F., et al. 2017,RMxAA, 53, 385Ferland, G. J., Korista, K. T., Verner, D. A., et al. 1998,PASP, 110, 761F¨orster Schreiber, N. M., Roussel, H., Sauvage, M., &Charmandaris, V. 2004, A&A, 419, 501 Gallagher, J. S., Bushouse, H., & Hunter, D. A. 1989, AJ,97, 700Hao, C.-N., Kennicutt, R. C., Johnson, B. D., et al. 2011,ApJ, 741, 124Hirashita, H., Buat, V., & Inoue, A. K. 2003, A&A, 410, 83Ho, L. C., & Keto, E. 2007, ApJ, 658, 314Houck, J. R., Roellig, T. L., Van Cleve, J., et al. 2004,Proc. SPIE, 5487, 62Houck, J. R., Weedman, D. W., Le Floc’h, E., & Hao, L.2007, ApJ, 671, 323Hunter, D. A., Elmegreen, B. G., & Ludka, B. C. 2010, AJ,139, 447Iwasawa, K., Sanders, D. B., Teng, S. H., et al. 2011, A&A,529, A106Izotov, Y. I., Lipovetsky, V. A., Chaffee, F. H., et al. 1997,ApJ, 476, 698Izotov, Y. I., & Thuan, T. X. 1999, ApJ, 511, 639Izotov, Y. I., & Thuan, T. X. 2004, ApJ, 616, 768Izotov, Y. I., Thuan, T. X., & Privon, G. 2012, MNRAS,427, 1229Jimmy, Tran, K.-V., Saintonge, A., et al. 2015, ApJ, 812, 98Kaneda, H., Onaka, T., Sakon, I., et al. 2008, ApJ, 684, 270Kelly, B. C. 2007, ApJ, 665, 1489Kennicutt Jr., R. C. 1998, ARA&A, 36, 189Kennicutt Jr., R. C., & Evans, N. J. 2012, ARA&A, 50, 531Kennicutt Jr., R. C., Hao, C.-N., Calzetti, D., et al. 2009,ApJ, 703, 1672Kewley, L. J., & Dopita, M. A. 2002, ApJS, 142, 35Kewley, L. J., & Ellison, S. L. 2008, ApJ, 681, 1183Kobulnicky, H. A., & Skillman, E. D. 1996, ApJ, 471, 211Kroupa, P., & Weidner, C. 2003, ApJ, 598, 1076LaMassa, S. M., Heckman, T. M., Ptak, A., et al. 2012,ApJ, 758, 1Lebouteiller, V., Barry, D. J., Goes, C., et al. 2015, ApJS,218, 21Lebouteiller, V., Barry, D. J., Spoon, H. W. W., et al. 2011,ApJS, 196, 8Leroy, A. K., Bolatto, A., Gordon, K., et al. 2011, ApJ,737, 12Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415Madden, S. C., Galliano, F., Jones, A. P., & Sauvage, M.2006, A&A, 446, 877Magdis, G. E., Daddi, E., B´ethermin, M., et al. 2012, ApJ,760, 6Maragkoudakis, A., Ivkovich, N., Peeters, E., et al. 2018,MNRAS, 481, 5370Markwardt, C. B. 2009, Astronomical Data AnalysisSoftware and Systems XVIII, 411, 2516