2.1. Dusty star-forming galaxies

One recent crucial development, which has opened up SPICA’s potential to probe the z > 5 Universe, was a series of discoveries finding that a significant fraction of the brightest submillimetre/millimetre sources in a random blank sky field corresponds to gravitationally lensed infrared-luminous galaxies at high redshift (except for nearby galaxies and bright AGN). The discovery of the Cosmic Eyelash galaxy at z = 2.3 (Swinbank et al. Reference Swinbank2010), which was the first of such super-bright (S ₈₇₀ > 100 mJy)Footnote ^c lensed infrared-luminous galaxies to be found, allowed a variety of multi-wavelength observations even with those observing facilities that normally do not have the sensitivity to probe beyond the low-redshift Universe.

Although this first discovery was serendipitous, wide-field surveys with Herschel, South Pole Telescope (SPT), Atacama Cosmology Telescope (ACT), and Planck quickly followed with more discoveries of similarly bright infrared-luminous galaxies (e.g., Negrello et al. Reference Negrello2010; Combes et al. Reference Combes2012; Vieira et al. Reference Vieira2013; Weiß et al. Reference Weiß2013; Marsden et al. Reference Marsden2014; Cañameras et al. Reference Cañameras2015; Harrington et al. Reference Harrington2016), a small number of which have turned out to be at z > 5. Due to lensing, these z > 5 galaxies are all substantially brighter than HDF 850.1 and AzTEC-3, so their redshifts were easily measured by blind CO searches. At the time of writing, the discoveries of ten such lensed infrared-luminous galaxies have been reported at z > 5 (Table 1; SPT0311-58 W and SPT0311-58 E are counted as one), with the highest-redshift galaxy at z = 6.90 (Strandet et al. Reference Strandet2017; Marrone et al. Reference Marrone2018). Even when corrected for lensing magnification (μ), many of these objects are hyper-luminous infrared galaxies (HyLIRGs; L _IR > 10¹³ L_⊙) but without any sign of a strong AGN, leading to their classification as DSFGs (Casey et al. Reference Casey, Narayanan and Cooray2014). Note that non-lensed z > 5 galaxies that are significantly more luminous than HDF 850.1 and AzTEC-3 are also being discovered (Table 1). With a variety of wide-field surveys being conducted/planned (see Section 4), the list of such HyLIRGs at z = 5–10, whether gravitationally lensed or intrinsically luminous, will grow rapidly over the coming years, providing excellent targets for SAFARI.

Table 1. Currently known infrared-luminous galaxies (μ L _IR ≳ 10¹³ L_⊙) at z > 5 (non-quasars)

^a Infrared luminosity L _IR (8–1 000 μm) without a lensing correction.

^b Magnification factor.

^c At 869 μm with ALMA.

^d At 880 μm with SMA.

^e At 850 μm with SCUBA-2.

^f At 920 μm with CSO/MUSIC.

^g At 890 μm with SMA.

^h J. Spilker 2018, private communication.

ⁱ L _IR (42–500 μm).

References: (1) Strandet et al. (Reference Strandet2017); (2) Marrone et al. (Reference Marrone2018); (3) Riechers et al. (Reference Riechers2013); (4) Cooray et al. (Reference Cooray2014); (5) Zavala et al. (Reference Zavala2018); (6) Strandet et al. (Reference Strandet2016); (7) Spilker et al. (Reference Spilker2016); (8) Vieira et al. (Reference Vieira2013); (9) Weiß et al. (Reference Weiß2013); (10) Gullberg et al. (Reference Gullberg2015); (11) Aravena et al. (Reference Aravena2016); (12) Ma et al. (Reference Ma2015); (13) Ma et al. (Reference Ma2016); (14) Combes et al. (Reference Combes2012); (15) Rawle et al. (Reference Rawle2014); (16) Asboth et al. (Reference Asboth2016); (17) Pavesi et al. (Reference Pavesi2018); (18) Riechers et al. (Reference Riechers2017); (19) Younger et al. (Reference Younger2007); (20) Smolčić et al. (Reference Smolčić2015); (21) Walter et al. (Reference Walter2012).

To illustrate the power of SAFARI, we show in Figure 1 the simulated SAFARI spectra of two gravitationally lensed infrared-luminous galaxies from Table 1, HLSJ0918 at z = 5.24 (Combes et al. Reference Combes2012; Rawle et al. Reference Rawle2014) and HFLS3 at z = 6.34 (Riechers et al. Reference Riechers2013; Cooray et al. Reference Cooray2014). These galaxies were discovered as Herschel sources showing red colours in the three SPIRE bands (S ₂₅₀ < S ₃₅₀ < S ₅₀₀), a technique that has proved to be effective for finding z > 4 DSFGs (e.g., Dowell et al. Reference Dowell2014). The figure clearly shows that SAFARI is capable of detecting main spectral features in the rest-frame mid-infrared at these redshifts if the infrared luminosities of target galaxies are > 10¹³ L_⊙. If the PAH features and fine-structure lines in these galaxies are as strong as those seen at lower redshift, SAFARI will be able to detect them clearly, and the measured PAH strengths can be used to estimate SFRs. Compared to other SFR indicators, PAH features have the advantage of being less vulnerable to dust extinction (e.g., compared to Hα) and being more luminous (e.g., compared to [Ne ii] 12.8 μm).

Figure 1. Simulated SAFARI spectra of HLSJ0918 (z = 5.24, μ = 9: Combes et al. Reference Combes2012; Rawle et al. Reference Rawle2014) and HFLS3 (z = 6.34, μ = 2: Riechers et al. Reference Riechers2013; Cooray et al. Reference Cooray2014) are shown in the left and right panels, respectively. The average local galaxy SED templates (Rieke et al. Reference Rieke, Alonso-Herrero, Weiner, Pérez-González, Blaylock, Donley and Marcillac2009) of L _IR = 10^11.75 and 10^12.50 L_⊙ were used, respectively, which produce good fits to the observed rest-frame far-infrared SEDs of these galaxies. The template SEDs were first scaled to the infrared luminosities without a lensing correction (μ L _IR in each panel) and then fit with PAHFIT (Smith et al. Reference Smith2007) with a pixel sampling of R = 600. These PAHFIT-produced model spectra were then redshifted and noise-added for corresponding integration times (T _int in each panel). Finally, the resultant spectra were resampled with R = 300 pixels. However, the effective resolution of these simulated spectra is less than R = 300 due to the low resolution (R ≈ 60–130) of the Spitzer/IRS data used by Rieke et al. (Reference Rieke, Alonso-Herrero, Weiner, Pérez-González, Blaylock, Donley and Marcillac2009) to build templates. Note that the actual mid-infrared spectra of these z > 5 galaxies may significantly differ from those of local LIRGs/ULIRGs (see Section 2.4 for more discussion).

PAH equivalent widths are also a powerful diagnostic for assessing the AGN contribution to the rest-frame mid-infrared continuum emission (e.g., Pope et al. Reference Pope2008; Riechers et al. Reference Riechers2014). Considering that many of the lensed infrared-luminous galaxies listed in Table 1 are HyLIRGs even intrinsically (i.e., when corrected for the lensing magnification), it is important to examine if they harbour luminous AGN and therefore exhibit smaller PAH equivalent widths. For the detection of AGN, especially those heavily obscured by dust, the rest-frame mid-infrared range is optimal as the AGN contribution becomes most conspicuous there. Other mid-infrared spectral features that can be used to detect the presence of AGN are high excitation lines such as [Ne V] 14.3/24.3 μm and [O IV] 25.9 μm, which can be used to estimate the AGN contribution and black hole accretion rates (e.g., Spinoglio et al. Reference Spinoglio2017).

Other prominent mid-infrared spectral features include atomic fine-structure lines such as [Ne ii]/[Ne iii] 12.8/15.6 μm and molecular hydrogen (H₂) lines such as 0–0 S(1)/0–0 S(3) 17.0/9.66 μm (some of these lines are not seen in Figure 1 because of their faintness and the low resolution of the template spectra). The [Ne ii]/[Ne iii] lines, for example, will serve as excellent indicators of SFRs and the hardness of ionising radiation (e.g., Thornley et al. Reference Thornley, Schreiber, Lutz, Genzel, Spoon, Kunze and Sternberg2000; Ho & Keto Reference Ho and Keto2007), while H₂ lines allow us to measure the temperature and mass of warm (T ≳ 100 K) molecular hydrogen gas directly (e.g., Rigopoulou et al. Reference Rigopoulou, Kunze, Lutz, Genzel and Moorwood2002; Higdon et al. Reference Higdon, Armus, Higdon, Soifer and Spoon2006).

Note that high-redshift HyLIRGs are likely more luminous in the rest-frame mid-infrared than the local ones, which will help SAFARI detections of submillimetre/millimetre-selected DSFGs like those listed in Table 1. This is because at z ≳ 1, many star-forming HyLIRGs/ULIRGs are spatially extended over kpc scales, exhibiting flatter and colder infrared spectral energy distributions (SEDs) that are more similar to those of local LIRGs [L _IR = 10¹¹ − 10¹² L_⊙; see Rujopakarn et al. (Reference Rujopakarn, Rieke, Weiner, Pérez-González, Rex, Walth and Kartaltepe2013) and references therein]. Indeed, the Herschel-observed far-infrared SEDs of HLSJ0918 and HFLS3 take shapes consistent with those of galaxies with much lower infrared luminosities (see the caption of Figure 1), supporting the validity of such an assumption.

2.2. UV-bright star-forming galaxies

At z ≳ 5, the majority of galaxies have been selected through robust optical (broad-band/narrow-band) colour selections and identified either as Lyman break galaxies (LBGs) or Lyman-alpha emitters (LAEs). LBGs and LAEs are inherently UV-bright star-forming galaxies because they are selected through the detections of the Lyman break at 912 Å and/or Lyα break/emission at 1 216 Å. Unlike DSFGs discussed above, which can be extremely faint in the rest-frame UV (e.g., HDF 850.1), LBGs/LAEs are less dust-obscured as populations, especially at z ≳ 5 where many of LBGs/LAEs are seen to exhibit extremely blue UV continuum slopes (e.g., Bouwens et al. Reference Bouwens2012; Dunlop et al. Reference Dunlop, McLure, Robertson, Ellis, Stark, Cirasuolo and de Ravel2012; Finkelstein et al. Reference Finkelstein2012; Jiang et al. Reference Jiang2013).

At z ∼ 3, Spitzer/IRS spectra exist for a small number of bright gravitationally lensed LBGs, such as MS1512-cB58 at z = 2.73 (Siana et al. Reference Siana, Teplitz, Chary, Colbert and Frayer2008) and the Cosmic Eye at z = 3.07 (Siana et al. Reference Siana2009), giving a glimpse of what the mid-infrared spectra of UV-selected star-forming galaxies look like. The mid-infrared spectra of these particular LBGs are similar to those of typical infrared-luminous galaxies like those in Figure 1, showing strong PAH features and resembling those of infrared-selected lensed galaxies at comparable redshift (Rigby et al. Reference Rigby2008). This is probably not surprising, considering that these LBGs are LIRGs in terms of their infrared luminosities and therefore are probably among the more infrared-luminous members of the LBG population. In fact, a significant fraction of z ⋍ 3 LBGs are thought to be infrared-luminous despite their rest-frame UV selection (e.g., Coppin et al. Reference Coppin2015; Koprowski et al. Reference Koprowski2016). A recent Herschel stacking analysis of about 22 000 z ⋍ 3 LBGs indicates that these galaxies are LIRGs on average (Álvarez-Márquez et al. Reference Álvarez-Márquez2016). It has also been shown that some of the z ⋍ 3 LBGs are even ULIRGs (e.g., Oteo et al. Reference Oteo2013; Magdis et al. Reference Magdis2017). Even at z ⋍ 7, bright LBGs are thought to be LIRGs on average (Bowler et al. Reference Bowler, Bourne, Dunlop, McLure and McLeod2018).

Because of the simple colour selection criteria, LBGs are known to constitute a heterogeneous sample of galaxies with a wide spectrum of physical properties, from dusty infrared-luminous galaxies to luminous LAEs with little dust extinction. One exciting prospect for SAFARI is that it will be able to detect the latter population (which likely dominates in number), making it possible to study both populations in a uniform way, using the same set of mid-infrared diagnostics.

In this context, particularly interesting are low-mass, low-metallicity, unreddened galaxies with strong emission lines at z ∼ 2, which may be better analogues of z ≳ 5 galaxies (e.g., Erb et al. Reference Erb, Pettini, Shapley, Steidel, Law and Reddy2010; Stark et al. Reference Stark2014). These galaxies may be similar to low-metallicity blue compact dwarfs (BCDs) in the local Universe (e.g., Watson et al. Reference Watson2011), and if so, their mid-infrared spectra are likely distinctly different from those of typical infrared-luminous galaxies shown in Figure 1. We will discuss the mid-infrared spectra of these local BCDs in Section 2.4.

2.3. Quasars/AGN

Compared to star-forming galaxies, quasars have much flatter infrared SEDs because of the power-law continuum produced by the central AGN. As a result, they are significantly brighter in the rest-frame mid-infrared and are easier to observe with SAFARI. Figure 2 shows the 100-μm flux-density distribution of 27 z > 5 Type-1 quasars (up to z = 6.4) based on the Herschel/Photodetector Array Camera and Spectrometer (PACS) photometry reported by Leipski et al. (Reference Leipski2014). Note that the PACS 100-μm band directly measures the source brightness in the wavelength range that SAFARI will cover. The measured 100-μm flux densities range from 2 to 12 mJy, indicating that SAFARI, with a 5σ continuum sensitivity of 0.7 mJy in 1 h, will be able to obtain high-quality spectra for these quasars quickly.

Figure 2. Herschel/PACS 100-μm photometry of 27 z = 5–6.4 Type-1 quasars reported by Leipski et al. (Reference Leipski2014). In comparison, the simulated 100- μm flux densities of HLSJ0918 and HFLS3 (see Figure 1) are also shown, as well as the sensitivity of SAFARI LR-mode (0.7 mJy at 100 μm, 5σ in 1 h). These z > 5 quasars are bright enough to be observable with SAFARI in under an hour (each), providing details about the dust composition and distribution of dust around their nuclei.

Mid-infrared spectra of low-redshift Type-1 AGN are often characterised by a power-law continuum, silicate emission/absorption features, and PAH emission features (e.g., Siebenmorgen et al. Reference Siebenmorgen, Haas, Krügel and Schulz2005; Hao et al. Reference Hao2005; Shi et al. Reference Shi2006, Reference Shi2007, Reference Shi, Rieke, Ogle, Jiang and Diamond-Stanic2009, Reference Shi, Rieke, Ogle, Su and Balog2014). The power-law continuum seen in the rest-frame mid-infrared is thought to be produced by the dusty torus around the central AGN (e.g., Leipski et al. Reference Leipski2014), allowing us to study the properties and geometry of the circumnuclear material. For example, the strengths of silicate emission/absorption features (at 9.7 and 18 μm) are thought to correlate (at least in the first order) with the orientation of the dusty torus (i.e., edge-on → absorption; face-on → emission) and can be used to infer the structure of the torus in the framework of unification models (e.g., Shi et al. Reference Shi2006). At z ≳ 6, some quasars are found to be deficient in hot dust, suggesting that their dusty tori are not fully developed or are even absent (Jiang et al. Reference Jiang2010; Leipski et al. Reference Leipski2014). SAFARI spectroscopy of z > 5 quasars therefore offers the possibility to investigate, through observations and modelling, the physical conditions and formation/evolution processes of AGN dusty tori.

Since the z > 5 quasars plotted in Figure 2 are so bright, the quality of SAFARI spectra will be high enough to examine the composition of dust grains. For example, using the Spitzer/IRS data for 93 AGN at z ≲ 0.5 that exhibit the 9.7 and 18 μm silicate emission features, Xie, Li, & Hao (Reference Xie, Li and Hao2017) have determined that 60 of these AGN spectra can be well reproduced by ‘astronomical silicates’, while 31 sources favour amorphous olivine (Mg_1.2Fe_0.8SiO₄) and two sources favour amorphous pyroxene (Mg_0.3Fe_0.7SiO₃). They also concluded that all sources require micron-sized dust grains, which are significantly larger than the submicron-sized dust grains found in the Galactic ISM. By measuring the central wavelength, width, and relative intensity of the two silicate features, SAFARI will allow us to infer the chemical composition and grain properties of the circumnuclear dust around AGN at z > 5 (see the companion paper by Fernandez-Ontiveros et al. (Reference Fernández-Ontiveros2017) for a further discussion of quasar mid-infrared spectra).

The PAH emission features, on the other hand, reveal star-forming activities in the quasar host galaxies (e.g., Shi et al. Reference Shi2007, Reference Shi, Rieke, Ogle, Jiang and Diamond-Stanic2009). Among the sample of Leipski et al. (Reference Leipski2014), there are seven z > 5 Type-1 quasars that have been detected at 500 μm. Although a significant fraction of z > 5 quasar far-infrared luminosities are thought to be produced by AGN [estimated to be 30–70% by Schneider et al. (Reference Schneider, Bianchi, Valiante, Risaliti and Salvadori2015) and Lyu, Rieke, & Alberts (Reference Lyu, Rieke and Alberts2016)], the infrared luminosities powered by star formation could still be larger than 10¹³ L_⊙ (Leipski et al. Reference Leipski2014). Such infrared luminosities are comparable to that of HLFS3 (z = 6.34) shown in Figure 1, suggesting that SAFARI will likely detect PAH emission features in many of these seven z > 5 quasars superposed on the power-law AGN continuum. Quasars with vigorously star-forming hosts may also allow us to examine the interplay between AGN and star formation at these early epochs.

2.4. Galaxies in the epoch of reionisation

As shown in Figure 1, SAFARI will be able to deliver good-quality rest-frame mid-infrared spectra for HyLIRGs at least up to z ∼ 6. The next question, therefore, is how much farther we can push SAFARI in redshift. The answer to this question depends on whether or not there exist HyLIRGs at z > 6 that are sufficiently massive and luminous to be detectable with SAFARI. Note that such high-redshift HyLIRGs are not explicitly included in some models of infrared-luminous galaxy evolution. For example, the model by Béthermin et al. (Reference Béthermin2017), one of the most advanced and up-to-date, applies a sharp SFR limit of < 1 000 M_⊙ yr⁻¹, excluding HyLIRGs like those listed in Table 1.

In this respect, the discoveries of HFLS3 at z = 6.34 (Riechers et al. Reference Riechers2013) and SPT0311-58 at z = 6.9 (Strandet et al. Reference Strandet2017; Marrone et al. Reference Marrone2018) are encouraging. The halo masses (M _h) of these high-redshift DSFGs have been estimated to be ≳ 10¹² M_⊙ (Marrone et al. Reference Marrone2018), and therefore their existence provides a proof that such massive infrared-luminous galaxies do exist at z ∼ 6–7, possibly marking the rare density peaks that would become present-day galaxy clusters and have a space density of only 10⁻³ − 10⁻⁴ times that of typical z ∼ 6 LBGs (Riechers et al. Reference Riechers2013).

Though rare, the existence of massive and luminous DSFGs is expected in overdense regions at z ∼ 6–7. For example, the simulation by Yajima et al. (Reference Yajima, Shlosman, Romano-Díaz and Nagamine2015) has shown that overdense regions evolve at a substantially accelerated pace at high redshift, being able to produce DSFGs at z ∼ 6 inside a halo with a mass of M _h ∼ 10¹² M_⊙. This simulation, however, failed to reproduce the observed infrared luminosity of HFLS3, falling short by a factor of about 10. One possible explanation is that HFLS3 is experiencing a powerful starburst that boosts the infrared luminosity. The same simulation also predicts the existence of L _IR ∼ 6 × 10¹¹ L_⊙ galaxies at z ∼ 10, and if their infrared luminosities are similarly enhanced by a strong starburst (i.e., by a factor of about 10), ULIRG-type galaxies may exist in some exceptional overdense regions even at z ∼ 10.

Figure 3 shows simulated 10-h spectra of z = 8 galaxies using the spectra/SEDs of the following three objects:

1. 1. HLSJ0918: The z = 5.24 gravitationally lensed infrared-luminous galaxy shown in Figure 1 (Combes et al. Reference Combes2012; Rawle et al. Reference Rawle2014).

2. 2. Haro 11: Local (D ≈ 90 Mpc) infrared-luminous (L _IR≈ 2 × 10¹¹ L_⊙) low-metallicity (Z ≈ 1/3 Z_⊙) BCD (e.g., Cormier et al. Reference Cormier2012; Lyu et al. Reference Lyu, Rieke and Alberts2016).

3. 3. II Zw 40: Another local (D ≈ 10 Mpc) low-metallicity (Z ≈ 1/5 Z_⊙) BCD with a significantly lower infrared luminosity of L _IR ≈ 3 × 10⁹ L_⊙ (see Consiglio et al. Reference Consiglio, Turner, Beck and Meier2016 and Kepley et al. Reference Kepley, Leroy, Johnson, Sandstrom and Chen2016 for recent ALMA studies and references). II Zw 40 is one of the two H II galaxies (along with I Zw 18) studied by Sargent & Searle (Reference Sargent and Searle1970), which have defined BCDs as a distinct class of galaxies.

Figure 3. SAFARI 10-h LR (R = 300) spectra for z = 8 galaxies simulated for the following three-types of galaxies: (a) HLSJ0918, a HyLIRG at z = 5.24 (see Figure 1 and Table 1); (b) Haro 11, a low-metallicity infrared-luminous local BCD; and (c) II Zw 40, another low-metallicity local BCD that is not infrared-luminous. For HLSJ0918, the L _{I R} = 10^11.75 L_⊙ LIRG SED from Rieke et al. (Reference Rieke, Alonso-Herrero, Weiner, Pérez-González, Blaylock, Donley and Marcillac2009) was used as in Figure 1, while for the two BCDs, the fully processed Spitzer/IRS low-resolution spectra were obtained from the Combined Atlas of Sources with Spitzer IRS Spectra (CASSIS; Lebouteiller et al. Reference Lebouteiller, Barry, Spoon, Bernard-Salas, Sloan, Houck and Weedman2011). The infrared luminosities of these SEDs have been scaled to 2 × 10¹³ L_⊙, comparable to the intrinsic luminosity of HFLS3. See the caption of Figure 1 for how these SAFARI spectra were simulated. The red lines show simulated H₂ emission lines (assumed to be unresolved) produced by 2 × 10¹⁰ M_⊙ of T = 200 K gas and 2 × 10⁸ M_⊙ of T = 1 000 K gas under the local thermodynamic equilibrium (LTE) assumption (an ortho-to-para ratio of 3:1 is also assumed). These H₂ lines are hardly visible in the original galaxy spectra.

The Spitzer/IRS spectra of the two BCDs were analysed by Hunt, Bianchi, & Maiolino (Reference Hunt, Bianchi and Maiolino2005) and Wu et al. (Reference Wu, Charmandaris, Hao, Brandl, Bernard-Salas, Spoon and Houck2006), while their broad-band SEDs (covering from near-infrared to submillimetre) were presented by Rémy-Ruyer et al. (Reference Rémy-Ruyer2015). As already mentioned in Section 2.2, these local BCDs are often thought to be good analogues of high-redshift low-metallicity galaxies (although the metallicities of actual z = 8 galaxies are likely even lower). The spectra of these BCDs were scaled up by assuming an infrared luminosity of 2 × 10¹³ L_⊙, comparable to the lensing-corrected luminosity of HFLS3.

As Figure 3 shows, 10-h integration SAFARI spectra will allow us to characterise the physical properties of such HyLIRGs at z ⋍ 8 in terms of the following characteristics: (i) PAH feature strengths; (ii) fine-structure line strengths; and (iii) underlying continuum shapes. For example, SAFARI will be able to test whether or not many of z > 5 galaxies are scaled-up versions of local low-metallicity BCDs. The mid-infrared spectra of low-metallicity BCD are distinctly different from those of normal infrared-luminous galaxies because of weak PAH features, strong high-excitation lines (e.g., [Ne iii] 15.5 μm and [S IV] 10.5 μm), and a sharply rising red continuum, as first reported by Madden et al. (Reference Madden, Galliano, Jones and Sauvage2006) based on ISO observations. Weak (or even absent) PAH features are a common characteristic of low-metallicity galaxies while strong high-excitation lines are likely due to a harder UV radiation field (Hunt et al. Reference Hunt, Bianchi and Maiolino2005; Wu et al. Reference Wu, Charmandaris, Hao, Brandl, Bernard-Salas, Spoon and Houck2006). The latter also explains the presence of the strong [O iii] 88-μm line (even more luminous than the [C II] 158-μm line) in low-metallicity dwarf galaxies, as recently observed by Herschel/PACS spectroscopy (Cormier et al. Reference Cormier2015). A powerful way to discriminate between low-metallicity BCDs, ‘normal’ (solar-metallicity) starburst galaxies, and AGN has also been derived from specific mid-infrared line ratios as presented by Fernández-Ontiveros et al. (Reference Fernández-Ontiveros, Spinoglio, Pereira-Santaella, Malkan, Andreani and Dasyra2016) (see their Figure 11) and Spinoglio et al. (Reference Spinoglio2017).

The recent detections of high-ionisation UV lines in high-redshift galaxies (e.g., Stark et al. Reference Stark2015a, Reference Stark2015b; Stark et al. Reference Stark2017; Mainali et al. Reference Mainali, Kollmeier, Stark, Simcoe, Walth, Newman and Miller2017) suggest that their mid-infrared spectra may also exhibit high-ionisation lines like those seen in these BCD spectra, and if so, that may support the idea that local BCDs are good analogues of high-redshift star-forming galaxies. A sharply increasing red continuum indicates a significantly warmer dust temperature (46.5 K in the case of Haro 11 by Lyu et al. Reference Lyu, Rieke and Alberts2016), and SAFARI will be effective for detecting such a warm-dust SED since its wavelength coverage extends to >20 μm in the rest-frame even at z = 8. The existence of such a warm-dust host galaxy has been suggested for z > 5 quasars based on their SED analysis (Lyu et al. Reference Lyu, Rieke and Alberts2016).

2.5. Molecular hydrogen (H₂) emission

The rest-frame mid-infrared spectral range is uniquely important, since it contains H₂ lines originating from the lowest energy levels (i.e., so-called H₂ pure-rotational linesFootnote ^d ), which allow us to measure the temperature and mass of the bulk of warm (T ≳ 100 K) molecular hydrogen gas in galaxies directly.

In local/low-redshift LIRGs and ULIRGs, the luminosities of the H₂ 0–0 S(1) line (which is normally one of the brightest pure-rotational lines) are typically around 0.005% of the total infrared luminosities [e.g., as estimated by Egami et al. (Reference Egami, Rieke, Fadda and Hines2006b) using the data from Rigopoulou et al. (Reference Rigopoulou, Kunze, Lutz, Genzel and Moorwood2002) and Higdon et al. (Reference Higdon, Armus, Higdon, Soifer and Spoon2006)]. With a 10-h integration, SAFARI’s 5σ line detection limit will be ≳ 10⁹ L_⊙ at z > 5, so this means that for a successful detection of the H₂ 0–0 S(1) line at z > 5, we would need a galaxy with a total infrared luminosity of > 2 × 10¹³ L_⊙, that is, HyLIRGs like those listed in Table 1.

By combining CLOUDY calculations (Ferland et al. Reference Ferland2013) with a zoom-in, high-resolution (⋍30 pc) numerical simulation, it is now possible to examine the physical conditions and internal structures of the inter-stellar medium (ISM) in high-redshift galaxies including molecular hydrogen gases (e.g., Vallini, Dayal, & Ferrara Reference Vallini, Dayal and Ferrara2012; Vallini et al. Reference Vallini, Gallerani, Ferrara and Baek2013, Reference Vallini, Gallerani, Ferrara, Pallottini and Yue2015; Pallottini et al. Reference Pallottini, Ferrara, Gallerani, Vallini, Maiolino and Salvadori2017a, Reference Pallottini, Ferrara, Bovino, Vallini, Gallerani, Maiolino and Salvadori2017b). So far, these simulations have explored the properties of average z ∼ 6 LBGs, and their H₂ line luminosities are predicted to be well below SAFARI’s detection limit (see Appendix A). For a successful SAFARI H₂ detection at such high redshift, we would therefore need a more massive galaxy undergoing a more violent H₂ heating process.

From the observations of the nearby and lower-redshift Universe, it is known that there exist galaxies that exhibit exceptionally strong H₂ emission. Examples include the local LIRG NGC 6240 (e.g., Lutz et al. Reference Lutz, Sturm, Genzel, Spoon, Moorwood, Netzer and Sternberg2003; Egami et al. Reference Egami, Neugebauer, Soifer, Matthews, Becklin and Ressler2006a; Armus et al. Reference Armus2006), the brightest cluster galaxy (BCG) in the centre of the X-ray-luminous cluster Zwicky 3146 (Z3146; z = 0.29; Egami et al. Reference Egami, Rieke, Fadda and Hines2006b), and the radio galaxy PKS1138-26 at z = 2.16 (the Spiderweb galaxy; Ogle et al. Reference Ogle, Davies, Appleton, Bertincourt, Seymour and Helou2012). The L(H₂ 0–0 S(1))/L _IR ratios of the first two galaxies are 0.03% and 0.25%, respectively, significantly larger than the typical value of 0.005% quoted above. No H₂ 0–0 S(1) measurement is available for the Spiderweb galaxy because of its high redshift, but the L(H₂ 0–0 S(3))/L _IR ratio is comparably high (0.4%). Such luminous H₂ emission lines are thought to be generated by mechanisms involving strong shocks, such as galaxy mergers (e.g., NGC 6240) and radio jets (e.g., the Spiderweb).

Note that some of the reported warm H₂ gas masses are exceptionally large, ∼ 10¹⁰ M_⊙ for the Z3146 BCG and ∼2 × 10¹⁰ M_⊙ for the radio galaxy 3C 433 at z = 0.1 (Ogle et al. Reference Ogle, Boulanger, Guillard, Evans, Antonucci, Appleton, Nesvadba and Leipski2010). However, their CO observations indicate that warm/cold H₂ mass ratios are very different between these two galaxies: ∼ 0.1 for the Zwicky 3146 BCG, which is a typical value for infrared-luminous galaxies, while > 3 for 3C 433, likely indicating an abnormally strong H₂ heating process.

Figure 4 shows the detectability of the brightest pure-rotational lines of three luminous H₂ emitters (NGC 6240, the Z3146 BCG, and the Spiderweb galaxy) toward high redshift. Although the H₂ 0–0 S(5) line of NGC 6240 would drop out of SAFARI detection at z ∼ 3, the H₂ 0–0 S(3) line of the Z3146 BCG would remain visible up to z ∼ 6, and the H₂ 0–0 S(3) line of the Spiderweb galaxy would stay well above the SAFARI detection limit even at z = 10. The figure also shows that if we assume an NGC 6240-like L (H₂ 0–0 S(1))/L _IR ratio (i.e., 0.03%), a HyLIRG with L _IR = 10¹³ L_⊙ will produce an H₂ 0–0 S(1) line detectable up to z ∼ 8, and with a Z3146 BCG-like ratio (i.e., 0.25%), it will be detectable beyond z = 10, just like the Spiderweb galaxy. The existence of these extreme H₂ emitters suggests that H₂ lines will likely serve as important probes for galaxies at high redshift, providing crucial observational constraints on theoretical models like the one presented for a z ⋍ 6 LBG in Appendix A.

Figure 4. Detectability of H₂ pure-rotational lines for three known extreme H₂ emitters: (1) the Spiderweb radio galaxy at z = 2.16 with L(0–0 S(3)) = 3.7 × 10¹⁰ L_⊙ (Ogle et al. Reference Ogle, Davies, Appleton, Bertincourt, Seymour and Helou2012); (2) Z3146 BCG at z = 0.29 with L(0–0 S(3)) = 1.6 × 10⁹ L_⊙ (Egami et al. Reference Egami, Neugebauer, Soifer, Matthews, Becklin and Ressler2006a); (3) NGC 6240 at z = 0.0245 with L(0–0 S(3)) = 3.4 × 10⁸ L_⊙ (Armus et al. Reference Armus2006). The brightest line was used for each case. With SAFARI’s line sensitivity, which is also plotted (10 h, 5σ), the Spiderweb galaxy would be visible beyond z = 10, while the Z3146 BGC would be visible up to z ∼ 6; NGC 6240, on the other hand, would drop out at z ∼ 3. Also shown are the visibilities of a HyLIRG (L _IR = 10¹³ L_⊙) through the 0–0 S(1) line assuming L _{0—0 S (1)}/L _IR = 0.25% (Z3146-like) and 0.03% (NGC 6240-like).

It should be emphasised that H₂ emission is sensitive only to warm (T ≳ 100 K) H₂ gas, meaning that it allows sampling of only a limited fraction (∼ 10–20%) of the total molecular gas mass in a typical galaxy. However, it is also possible to estimate the total H₂ gas mass from the measured warm H₂ gas mass by making some assumptions. For example, Togi & Smith (Reference Togi and Smith2016) recently proposed such a method, estimating the total H₂ gas mass from the observations of multiple H₂ emission lines alone, assuming a continuous power-law distribution of rotational temperatures down to a certain cutoff value. For a sample of local galaxies with reliable CO-based molecular masses, this method has been shown to produce the total molecular gas mass within a factor of 2 of those derived from CO when a cut-off temperature of around 50 K is adopted. Though indirect and dependent on some assumptions, methods like this have the potential to provide useful estimates for total H₂ gas masses, especially for low-metallicity galaxies, for which CO-/dust-based methods are known to underpredict molecular gas mass by a factor of approximately 100 locally, possibly due to the presence of CO-dark H₂ gas (Wolfire, Hollenbach, & McKee Reference Wolfire, Hollenbach and McKee2010; Togi & Smith Reference Togi and Smith2016).

Taking these known extreme H₂ emitters as a guide, we also simulated the spectra of H₂ lines in Figure 3, by making the following two assumptions: (1) H₂ level populations are fully thermalised (i.e., in the LTE); and (2) the galaxy contains two warm H₂ gas components, one with a gas mass of 2 × 10¹⁰ M_⊙ and a gas temperature of T = 200 K and the other with a gas mass of 2 × 10⁸ M_⊙ and a gas temperature of T = 1 000 K. An ortho-to-para ratio of 3:1 is also assumed. Such a two-component LTE model is known to produce good fits to the excitation diagrams of H₂ pure-rotational lines for lower-redshift galaxies (e.g., Higdon et al. Reference Higdon, Armus, Higdon, Soifer and Spoon2006), although this should probably be taken as a simple and effective parameterisation of more complex underlying gas temperature and mass distributions.

The red lines shown in Figure 3 indicate the H₂ emission lines produced by such a model. These simulated H₂ lines have luminosities of 0.5–1 × 10¹⁰ L_⊙, exceeding SAFARI’s 10-h 5σ line-flux limit of 3 × 10⁹ L_⊙ at z = 8. If detected, such luminous H₂ lines would indicate the existence of a large warm H₂ gas reservoir, as well as some mechanism that heats it (e.g., shocks), possibly marking the sites of galaxy formation/assembly.

3.1. First objects: Current picture

Although first stars and galaxies are yet to be observed, they have been a major focus of theoretical studies over the years [see Ciardi & Ferrara (Reference Ciardi and Ferrara2005), Bromm & Yoshida (Reference Bromm and Yoshida2011), Yoshida, Hosokawa, & Omukai (Reference Yoshida, Hosokawa and Omukai2012), Bromm (Reference Bromm2013), Greif (Reference Greif2015), and Barkana (Reference Barkana2016) for review]. In the framework of the standard ΛCDM model, we expect the first (i.e., Pop III) stars to form in dark matter (DM) minihalos of around 10⁶ M_⊙ at redshifts z ⋍ 20–30, cooling via H₂ molecular lines (Haiman, Thoul, & Loeb Reference Haiman, Thoul and Loeb1996; Tegmark et al. Reference Tegmark, Silk, Rees, Blanchard, Abel and Palla1997; Yoshida et al. Reference Yoshida, Abel, Hernquist and Sugiyama2003). The first stars formed in such a metal-free environment are believed to be quite massive (> 100 M_⊙; e.g., Hirano et al. Reference Hirano, Hosokawa, Yoshida, Omukai and Yorke2015), and would emit strong H₂ -dissociating UV radiation (e.g., Omukai & Nishi Reference Omukai and Nishi1999) and produce powerful supernova explosions (Bromm, Yoshida, & Hernquist Reference Bromm, Yoshida and Hernquist2003), essentially shutting off subsequent star formation. For this reason, these minihalos are not regarded as ‘first galaxies’ although they are the sites of the first star formation. The next generation of star formation will then take place in more massive halos (∼ 10⁸ M_⊙) collapsing at z ∼ 10, whose virial temperature is high enough (> 10⁴ K) to sustain cooling due to atomic hydrogen (e.g., Oh & Haiman Reference Haiman2002). These so-called atomic cooling halos hosting the second generation of stars are often considered as ‘first galaxies’ (Bromm & Yoshida Reference Bromm and Yoshida2011).

Note that according to this current standard picture, first galaxies are not necessarily metal-free (Pop III), which is often taken as the observational definition of the first galaxies. In fact, ‘This popular definition of a first galaxy may be misleading and may render any attempts to find first galaxies futile from the very outset’ (Bromm & Yoshida Reference Bromm and Yoshida2011). This is because it is difficult to prevent minihalos, that is, the building blocks of first galaxies, from forming massive Pop III stars and chemically enriching their surroundings through SNe explosions. In other words, to produce genuine Pop III galaxies, it is necessary to inhibit star formation in the progenitor mini-halos by suppressing the formation of molecular hydrogen in them. This would require H₂ -dissociating Lyman–Werner (LW) background radiation in the Far-UV (11.2–13.6-eV photons) and the source of such radiation before the formation of Pop III galaxies. This leads to a scenario in which the first galaxies that appeared in the Universe (in the chronological sense) were mostly Pop II galaxies; Pop III galaxies would appear subsequently in underdense regions where the star formation in minihalos was suppressed by radiation emitted by stars/galaxies formed earlier in overdense regions. For this reason, Pop III galaxies may be considered as the second-generation galaxies containing first-generation stars (e.g., Johnson, Greif, & Bromm Reference Johnson, Greif and Bromm2008; Trenti & Stiavelli Reference Trenti and Stiavelli2009; Johnson Reference Johnson2010; Stiavelli & Trenti Reference Stiavelli and Trenti2010; Johnson, Dalla Vecchia, & Khochfar Reference Johnson, Dalla Vecchia and Khochfar2013).

Adopting this theoretical picture as the baseline, we will discuss below the rest-frame mid-infrared spectral signatures of massive (M _h ∼ 10¹¹ – 10¹² M_⊙) forming galaxies containing low-metallicity or pristine (i.e., metal-free) gas clouds. By ‘forming’, we denote galaxies that are yet to form stars, meaning that the source of line luminosities is the gravitational energy released by the contraction of clouds under their self-gravity, rather than stellar radiation.

Production of the first dust by Pop III SNe would be another important scientific topic that could be uniquely addressed by rest-frame mid-infrared spectroscopy. However, considering that even larger uncertainties are involved in such a discussion, we limit ourselves here to providing a qualitative overview in Appendix B, deferring a more quantitative analysis to a forthcoming paper (R. Schneider et al., in preparation).

3.2. Massive forming galaxies

3.2.1. Population II

As already discussed above, formation of Pop II galaxies may precede that of Pop III galaxies in the history of the Universe. When the gas metallicity exceeds a certain thresholdFootnote ^e , gas cooling will be dominated by several key fine-structure lines, such as [C II] 158 μm and [O I] 63 μm in the far-infrared (Bromm & Loeb Reference Bromm and Loeb2003) and [Si II] 34.8 μm and [Fe II] 25.99/35.35 μm in the mid-infrared (Santoro & Shull Reference Santoro and Shull2006) if we assume that C, Si, and Fe atoms are photoionised by ambient UV radiation. The former far-infrared lines, when redshifted, can be studied with submillimetre/millimetre telescopes on the ground (e.g., ALMA), but the detection of the latter mid-infrared lines requires a space far-infrared telescope such as SPICA Footnote ^f .

For SAFARI to have any chance of detection, the target object must be exceptionally luminous and massive. As a maximal model, we consider here a Pop-II forming galaxy that has a gas mass (M _gas) of 10¹¹ M_⊙. The corresponding halo mass (M _h) would be ∼ 10¹² M_⊙, and such a massive halo has a comoving number density of ∼ 10⁻⁵ Mpc⁻³ at z ∼ 6. The resultant gas mass fraction would be ∼ 10%, which is still well below the cosmic baryon fraction of ∼ 0.19 (f_b = Ω _b / Ω _c ; Planck Collaboration et al. 2016). Such a classical model of a massive forming galaxy containing a uniform gas is obviously an over-simplification, but it serves the purpose of illustrating the parameter space SAFARI could potentially probe.

Figure 5 plots the expected luminosities of key cooling lines from such a massive forming galaxy assuming a metallicity of Z = 0.01 Z_⊙. Figure 5(a) shows that if the gas is sufficiently warm (T ≳ 200 K) and dense (n _H ∼ 10⁵–10⁶ cm⁻³) to thermalise all these lines (i.e., in LTE), the mid-infrared fine-structure lines such as the [Fe II] 25.99/35.35- μm and [Si II] 34.8- μm lines can be quite luminous (≳ 10¹¹ L_⊙) and detectable with SAFARI to high redshift (Note, however, that these lines will be redshifted out of the SAFARI window at z = 7.8/5.5, and 5.6, respectively). On the other hand, if the typical gas density is more like ∼ 10³ cm⁻³, these mid-infrared fine-structure lines will become undetectable beyond z ∼ 5 even for such a massive forming galaxy [Figure 5(b)].

Figure 5. Luminosities of H₂ lines (0–0 S(1) 17 μm and 0–0 S(0) 28 μm) and fine-structure lines ([Fe II] 25.99/35.35 μm, [Si II] 34.8 μm, [O I] 63 μm, and [C II] 158 μm) calculated for a low-metallicity (Z = 0.01 M_⊙) forming galaxy with a gas mass of 10¹¹ M_⊙. The left panel (a) assumes a gas density high enough for these lines to be thermalised (i.e., in LTE) while the right panel (b) corresponds to the case with a gas density of 10³ cm⁻³. A molecular fraction of 2 × 10⁻⁴ was assumed for both cases. The former is similar to the calculation presented by Santoro & Shull (Reference Santoro and Shull2006) (see their Figure 11 for a similar model with a gas mass of 10⁸ M_⊙). The horizontal dotted lines indicate the nominal 5σ detection limit of SAFARI with a 10-h on-source integration time for z = 3, 5, and 10. The LTE case on the left suggests that if the gas is sufficiently warm (T ≳ 200 K), fine-structure lines like [Fe II] 25.99/35.35 μm and [Si II] 34.8 μm can be quite luminous (note, however, that these lines will be redshifted out of the SAFARI window at z = 7.8/5.5, and 5.6, respectively).

As discussed by Santoro & Shull (Reference Santoro and Shull2006), H₂ lines in these models are significantly fainter when compared to the fine-structure lines. This is because the molecular fraction was assumed to be 2 × 10⁻⁴ here, resulting in a total H₂ gas mass of only 2 × 10⁷ M_⊙. As already shown in Figure 3, for SAFARI to detect H₂ lines at high redshift, the corresponding H₂ gas mass needs to be ≳ 10¹⁰ M_⊙. Note that the luminosities of H₂ lines, as well as that of [C II] 158 μm, do not change much between the two cases because a gas density of 10³ cm⁻³ is close to the critical densities of these transitions.

3.3. Pop III objects at lower redshift

Although detecting genuine Pop III objects at high redshift will likely require strong boosting of H₂ line luminosities by some mechanism as well as lensing amplification, SAFARI may be able to probe the properties of such objects through the observations of similarly metal-poor objects that may exist at lower redshift. For example, a number of studies have suggested that Pop III star formation may continue towards low redshift and maybe even down to z ∼ 3 (e.g., Scannapieco, Schneider, & Ferrara Reference Scannapieco, Schneider and Ferrara2003; Jimenez & Haiman Reference Jimenez and Haiman2006; Schneider et al. Reference Schneider, Salvaterra, Ferrara and Ciardi2006a; Tornatore, Ferrara, & Schneider Reference Tornatore, Ferrara and Schneider2007; Ricotti, Gnedin, & Shull Reference Ricotti, Gnedin and Shull2008; Trenti & Stiavelli 887 Reference Trenti and Stiavelli2009; Johnson Reference Johnson2010; Johnson et al. Reference Johnson, Dalla Vecchia and Khochfar2013; Pallottini et al. Reference Pallottini, Ferrara, Gallerani, Salvadori and D’Odorico2014, Reference Pallottini, Gallerani, Ferrara, Yue, Vallini, Maiolino and Feruglio2015a). According to the recent Renaissance Simulations (Xu et al. Reference Xu, Norman, O’shea and Wise2016), only 6% of the volume and 13% of the gas mass are enriched to [Z/H] > −4 at z = 7.6 in the comoving survey volume of 220 Mpc³, indicating that there is a large amount of pristine gas available for Pop III star formation at z > 7.6 (although much of it likely resides in low-density diffuse IGM). Observationally, however, there has been no secure identification of a Pop III galaxy so far. Although there are indications of massive, low-metallicity stars in the nearby lowest-metallicity galaxy I Zw 18 (Kehrig et al. Reference Kehrig, Vílchez, Pérez-Montero, Iglesias-Páramo, Brinchmann, Kunth, Durret and Bayo2015), a clear detection of Pop III objects is still missing. Note that, at z = 3, the line-luminosity detection limit of SAFARI will be lower by an order of magnitude (∼3 × 10⁸ L_⊙, 5σ in 10 h) compared to that at z = 8, potentially enhancing the probability of detecting Pop III objects.

4.1. Surface density of massive halos at z > 5

Whether we try to detect hyper-luminous galaxies like HFLS3/SPT0311-58 (Section 2.4) or a massive Pop-II forming galaxy (Section 3.2.1), we will be searching for objects residing in massive halos with M _h ≥ 10¹² M_⊙. Figure 6 shows the cumulative surface density of ≥ 10¹² M_⊙ halos per 1 000 deg² (the black line). Together, we also plot the surface densities of ≥10¹¹ M_⊙ halos gravitationally lensed by a factor of ≥10 × (the red line) and ≥10¹⁰ M_⊙ halos lensed by a factor of ≥100 × (the blue line). If we make a simple assumption that various physical properties of galaxies hosted by these halos (e.g., SFR, M _*) roughly scale linearly with the halo mass at z > 5 (i.e., maintaining a constant mass-to-light ratioFootnote ⁱ ), these three populations of halos would contain galaxies that have comparable apparent brightnesses. The cumulative surface density of the three halo populations combined (the thick grey line) is about 2 200 at z ≥ 7, 280 at z ≥ 8, 50 at z ≥ 9, and 12 at z ≥ 10 (per 1 000 deg²). The figure also shows that lensed populations would start to dominate in number at z ≳ 8.5, especially ≥ 10¹⁰ M_⊙ halos magnified by a factor of ≥100 ×.

Figure 6. Cumulative surface densities of DM halos per 1 000 deg² as predicted by the standard ΛCDM model. The black line indicates the cumulative surface density of ≥ 10¹² M_⊙ halos down to given redshifts while the blue and red lines plot the corresponding numbers for ≥ 10¹¹ M_⊙ halos gravitationally lensed by a factor of ≥ 10 × (μ ≥ 10) and for ≥ 10¹⁰ M_⊙ halos lensed by a factor of ≥ 100 × (μ ≥ 100), respectively. According to the calculation by Zackrisson et al. (Reference Zackrisson, González, Eriksson, Asadi, Safranek-Shrader, Trenti and Inoue2015), when the source redshift is above ∼ 5, the corresponding lensing probability is roughly constant and ∼ 10⁻⁴ for μ ≥ 10 and ∼ 10⁻⁶ for μ ≥ 100 along an average line of sight. These probabilities have been multiplied with the ≥ 10¹¹ and ≥ 10¹⁰ M_⊙ halo surface densities. The thick grey line plots the total surface density of the three halo populations combined while the thick grey dashed line plots 3% of the total number, comparable to the starburst fraction estimated at lower redshift (e.g., Béthermin et al. Reference Béthermin2017). The halo comoving number density was computed with the Press–Schechter formalism (Press & Schechter Reference Press and Schechter1974). The horizontal dashed line corresponds to N = 1.

Note, however, that translating the surface density of halos into that of HyLIRGs (μ L > 10¹³ L_⊙) or massive forming galaxies is not trivial due to a number of astrophysical processes involved. In this sense, Figure 6 sets the upper limit on the number of possible detections (i.e., there cannot be more galaxies than there are halos). In practice, we may assume that only a fraction of these halos hosts luminous objects that are observable with SAFARI. As an illustration, we also plot a line denoting 3% of the total halo number, comparable to a starburst fraction assumed by some models (e.g., Béthermin et al. Reference Béthermin2017). Although luminous objects may be more abundant at high redshift due to increased star/structure-formation activities, these simple calculations indicate that it will likely be necessary to survey hundreds/thousands of square degrees to find a handful of high-redshift objects that are luminous enough for SAFARI to observe, which is consistent with the outcomes of the wide-field Herschel /SPT surveys so far.

4.2. Submillimetre/millimetre surveys

Wide-field submillimetre/millimetre surveys have proven to be extremely effective in finding gravitationally lensed infrared-luminous galaxies at high redshift (e.g., Negrello et al. Reference Negrello2010; Vieira 950 et al. Reference Vieira2013; Weiß et al. Reference Weiß2013). This is because (1) the apparent brightnesses of infrared-luminous galaxies do not fade much in the submillimetre/millimetre to high redshift (up to z ∼ 10 at ∼1 mm) due to a strong negative K correction and (2) foreground galaxies are faint in the submillimetre/millimetre unless they contain strong AGN or they are at really low redshift. As a result, it is straightforward to identify lensed infrared-luminous galaxies at high redshift by just inspecting the brightest sources in the submillimetre/millimetre, 20–30% of which are typically lensed galaxies at z ≳ 1 (Negrello et al. Reference Negrello2010).

Recent results from Herschel /SPIRE wide-field surveys show that the surface density of bright (S ₅₀₀ > 100 mJy) lensed infrared-luminous galaxies ranges from 0.13 deg⁻² (80 sources over 600 deg², Negrello et al. Reference Negrello2017) to 0.21 deg⁻² (77 sources over 372 deg², Nayyeri et al. Reference Nayyeri2016). The surface density of SPT-selected galaxies has been shown to be consistent with that of these Herschel /SPIRE-selected galaxies (Mocanu et al. Reference Mocanu2013). These low (<1 deg⁻²) surface densities clearly indicate that wide submillimetre/millimetre surveys covering hundreds/thousands of square degrees are needed to produce a large sample of such lensed infrared-luminous galaxies, only a limited fraction of which will be at z > 5.

The depth of Herschel /SPIRE wide surveys matches well with the spectroscopic sensitivity of SAFARI. HFLS3 (z = 6.34) shown in Figure 1 has a SPIRE 500- μm flux density of 47 mJy, while the corresponding 5σ confusion limit is approximately 35 mJy (Nguyen et al. Reference Nguyen2010). The figure indicates that it will be difficult for SAFARI to obtain good-quality spectra for sources much fainter than HFLS3. In this sense, SPICA will be a well-matched spectroscopic follow-up mission for Herschel -selected z > 5 galaxies.

Note also that most of these submillimetre/millimetre-detected lensed galaxies are located at moderate redshift (z = 1–4). Although these lower-redshift galaxies will make excellent SAFARI targets in general (with a typical magnification factor of 10, integration times will be reduced by a factor of 100), z > 5 galaxies are much more scarce. Compared to the Herschel surveys, the SPT survey has been more successful in finding z > 5 galaxies (see (see Table 1) presumably because the effect of the negative K correction extends towards higher redshift in the millimetre. This suggests that future/ongoing millimetre surveys offer great potential for finding bright targets for SAFARI at the highest redshifts (e.g., Advanced ACT, SPT3G, as well as new cameras on the APEX, IRAM 30-m, and LMT telescopes).

A complementary approach would be to search for lensed infrared-luminous galaxies in the fields of massive lensing clusters. In fact, HLSJ0918, one of the most infrared-luminous galaxies at z ≳ 5 identified so far (see Table 1 and Figure 1) was discovered by the Herschel Lensing Survey (HLS) targeting such lensing cluster fieldsFootnote ^j (Egami et al. Reference Egami2010; Combes et al. Reference Combes2012; Rawle et al. Reference Rawle2016).

4.3. Lyα survey

Although submillimetre/millimetre surveys have been quite successful so far for finding luminous (often lensed) infrared-luminous galaxies at high redshift (up to z = 6.9), the fraction of such an infrared-luminous galaxy population will likely decrease at higher redshift, where the metallicities of galaxies are significantly lower on average. For the detection of less dusty and therefore UV-bright galaxies, wide-field Lyα emitter surveys will be effective. Although Lyα emission can be strongly suppressed at z > 6 by increasingly neutral IGM, the line may survive if the Lyα-emitting galaxy is located in a large H ii bubble (e.g., Cen & Haiman Reference Cen and Haiman2000; Haiman Reference Haiman2002). In fact, the tentative detection of Lyα emission from the z = 9.1 galaxy recently reported by Hashimoto et al. (Reference Hashimoto2018b) supports this idea. In the next subsection (Section 4.4), we will discuss how we can find such H ii bubbles with the Square Kilometer Array (SKA).

Search for strong Lyα emitters will be particularly powerful for finding massive forming galaxies discussed in Section 3.2. These objects will have little continuum emission before stars are formed, so strong emission lines like Lyα will be the only available tracers for such objects. As already discussed in Section 3.2.2, luminous Pop III H₂ emitters are likely luminous Lyα emitters as well.

In this sense, one particularly interesting recent example is CR7 at z = 6.6 (Sobral et al. Reference Sobral, Matthee, Darvish, Schaerer, Mobasher, Röttgering, Santos and Hemmati2015). Its exceptionally bright but narrow Lyα line (2 × 10¹⁰ L_⊙), coupled with the detection of a strong He II 1640 Å line and a lack of any metal lines, was originally taken as the sign that this object might harbour a Pop III stellar population. Such a Pop III scenario, however, encountered many difficulties because of the extreme conditions required for the underlying Pop III stellar population (e.g., Pallottini et al. Reference Pallottini2015b; Yajima & Khochfar Reference Yajima and Khochfar2017; Visbal, Bryan, & Haiman Reference Visbal, Bryan and Haiman2017) as well as of the possible presence of strong [O III] 4959/5007 Å lines inferred from the Spitzer /IRAC photometric data (Bowler et al. Reference Bowler, McLure, Dunlop, McLeod, Stanway, Eldridge and Jarvis2017). Although other interpretations were also put forth for the nature of CR7, such as a direct collapse black hole (DCBH, Pallottini et al. Reference Pallottini2015b; Agarwal et al. Reference Agarwal, Johnson, Zackrisson, Labbe, van den Bosch, Natarajan and Khochfar2016; Dijkstra, Gronke, & Sobral Reference Dijkstra, Gronke and Sobral2016; Smith, Bromm, & Loeb Reference Smith, Bromm and Loeb2016; Agarwal et al. Reference Agarwal, Johnson, Khochfar, Pellegrini, Rydberg, Klessen and Oesch2017; Pacucci et al. Reference Pacucci, Pallottini, Ferrara and Gallerani2017), the recent ALMA observations by Matthee et al. (Reference Matthee2017) with the detection of the [C II] 158-μm line have concluded that CR7 is a system that is undergoing the build-up process of a central galaxy through complex accretion of star-forming satellites. The original He ii line detection was also not reproduced by the reanalysis of the same data by Shibuya et al. (Reference Shibuya2018) although a re-analysis done by the CR7 discovery team still detects the line (Sobral et al. Reference Sobral2017), leaving the situation unclear.

Although CR7 is not likely to be a Pop III galaxy, identifying similarly luminous LAEs may prove to be the key to finding luminous H₂ emitters at high redshift. The most luminous H₂ emitters are those hosted by H i atomic cooling halos, which would also emit comparably luminous Lyα if the line is not significantly absorbed by the IGM (see Section 3.2.2). Since the typical line detection limit of current z ∼ 6–7 narrow-band LAE surveys is ∼ 2 × 10⁹ L_⊙ (e.g. Ouchi et al. Reference Ouchi2018), if the corresponding H₂ line luminosities are on the same order, SAFARI should be able to detect such H₂ lines in ∼10 h (5σ). Without having the capability to conduct a sensitive wide-field H₂ line survey in the mid-/far-infrared directly, Lyα -based surveys will likely play an important role for finding strong H₂ emitters (and strong mid-infrared line emitters in general) at high redshift.

4.4. SKA survey

In the coming decade, the SKA will start to provide tomographic views of the high-redshift Universe through the 21-cm emission line of neutral hydrogen (e.g., Koopmans et al. Reference Koopmans2015). The wide-field SKA map will be ideal for identifying large bubbles of ionised (H ii) gas at high redshift, marking the concentrations of star-forming galaxies in the early Universe. Multi-wavelength imaging observations targeting these H ii bubbles and the area around them will likely provide interesting targets for SAFARI.

Figure 7 shows halos with M _h > 10¹¹ M_⊙ (the blue dots) superimposed on the background ionisation field from the simulated tomographic map at z = 8, both projected from a 100-Mpc thick slice (see Sobacchi & Mesinger Reference Sobacchi and Mesinger2014 for the methodology of the simulation). The figure shows that halos are significantly clustered towards the centre of H ii bubbles (seen as bright white spots) because these halos are the sites of star formation and therefore the sources of ionising radiation. These H ii bubbles have a size of 20–30 Mpc (comoving), corresponding to approximately 10 arcmin on the sky.

Figure 7. Distribution of dark halos with M _h ≥ 10¹¹ M_⊙ at z = 8 (blue dots) superimposed on the background ionisation field (grey scale) taken from the simulated tomographic map at z = 8 shown over an area of 500 Mpc (comoving) on a side, which corresponds to 3.2 deg. This map was smoothed over a scale of 30 Mpc. The darker (brighter) regions correspond to more neutral (ionised) regions. Both distributions were projected from a 100-Mpc thick slice at z = 8. The map of ionisation field was produced based on the FULL model of Sobacchi & Mesinger (Reference Sobacchi and Mesinger2014).

Such an SKA H i map can be used in two ways. First, we can look for hyper-luminous galaxies and massive Pop-II forming galaxies inside large H ii bubbles, which mark rare high-σ peaks of the background cosmic density field and harbour many star-forming galaxies. Second, we can search for massive Pop-III forming galaxies around H ii bubbles. The formation of massive Pop III galaxies would require not only pristine metal-free environments (i.e., outside H ii bubbles) but also strong UV radiation to suppress H₂/H i cooling and therefore star formation until minihalos are assembled into galaxy-size halos (i.e., around H ii bubbles; see the discussion in Section 3.1).

4.5. Next-generation optical/near-infrared wide-field surveys

If accepted as the ESA M5 mission, SPICA will be launched around 2030. From the SPICA high-redshift science point of view, this lead time will allow us to take advantage of the next-generation optical/near-infrared wide-field surveys for target selection, such as Large Synoptic Survey Telescope (LSST), Euclid, and Wide Field Infrared Survey Telescope (WFIRST). For example, the High Latitude Survey of WFIRST is expected to detect around 100 bright (m _{160, AB} < 26 mag) galaxies at z > 10 over a survey area of 2 200 deg² using the continuum drop-out technique (Spergel et al. Reference Spergel2015). Galaxies at such high redshift represent rare peaks in the cosmic density field, marking the underlying regions of large-scale density enhancement, where we can expect the existence of many more objects at the same redshift (e.g., Barkana Reference Barkana2016). z > 10 objects discovered by WFIRST (as well as Euclid) will be strong rest-frame UV-continuum sources, which are likely unobscured star-forming galaxies, but the regions marked by these sources would likely harbour other types of high-redshift objects, such as infrared-luminous (dusty/dust-obscured) galaxies and massive forming galaxies. Grism slitless spectroscopic surveys with Euclid and WFIRST will also discover high-redshift Lyα emitters ((e.g., Bagley et al. Reference Bagley2017; Marchetti, Serjeant, & Vaccari Reference Marchetti, Serjeant and Vaccari2017), some of which may turn out to be genuine Pop III galaxies. These wide-field surveys will also be useful for finding exceptionally bright gravitationally lensed galaxies, which are expected to dominate the WFIRST sample of bright z > 10 galaxies (Spergel et al. Reference Spergel2015).

Another exciting possibility is that these next-generation deep wide-field surveys may detect directly the explosions of massive Pop III SNe, so-called pair-instability SNe (PISNe; see see Appendix B), at z ≲ 7–8 with LSST, out to z ∼ 15–20 with WFIRST, and out to z ≳ 30 with JWSTFootnote ^k (Whalen et al. Reference Whalen2013a; see also Wang et al. Reference Wang2017). These observations will catch PISNe in the initial UV-bright unobscured phase, and if they are detected and studied, they will help us extend our understanding to the subsequent dust-enshrouded phase, which is directly relevant to SAFARI observations as further discussed in Appendix B. Furthermore, it is possible that some supermassive stars (∼ 10⁵ M_⊙) may explode as thermonuclear SNe with energies of ∼ 10⁵⁵ erg, 100 times higher than those of Pop III PISNe, making them the most energetic explosions in the Universe (Whalen et al. Reference Whalen2013b). Such energetic events, if they exist, may also produce objects that can be studied with SAFARI.

4.6. SPICA/SMI surveys

SPICA /SMI (Kaneda et al. Reference Kaneda2017; Roelfsema et al. Reference Roelfsema2018) has the ability to conduct low-resolution (LR; R = 50–120) multi-slit prism spectroscopy covering 17–36 μm over a field of view of 12 arcmin × 10 arcmin. It also has a slit-viewer camera (CAM), which can perform 34 μm imaging over the same field of view. These SMI modes will allow wide-field survey programs, and some preliminary ideas have been presented in the companion papers, such as the SMI/LR deep (1 deg²) and wide (10 deg²) spectroscopic surveys (Kaneda et al. Reference Kaneda2017) and SMI/CAM ultra-deep (0.2 deg²), deep (1 deg²), and shallow (600 deg²) imaging surveys (Gruppioni et al. Reference Gruppioni2017).

For the purpose of finding bright z > 5 targets for SAFARI, these SMI surveys are particularly effective for detecting AGN-dominated galaxies, which can be substantially brighter than star-forming galaxies in the rest-frame mid-infrared. At z > 5, SMI will observe <6 μm in the rest-frame, where the SEDs of typical star-forming galaxies exhibit a broad trough between a stellar continuum peaking in the near-infrared and a dust continuum peaking in the far-infrared (see Gruppioni et al. Reference Gruppioni2017). In the case of AGN-dominated galaxies, however, a bright power-law AGN continuum could fill this SED trough and could even overwhelm the light from the host galaxy all together. As a result, the SMI-selected sample of z > 5 HyLIRGs is expected to be dominated by galaxies with AGNFootnote ^l (Gruppioni et al. Reference Gruppioni2017; Kaneda et al. Reference Kaneda2017). Such a sample will be particularly useful for studying the interplay between star formation and black-hole accretion (Spinoglio et al. 1157 Reference Spinoglio2017; Gruppioni et al. Reference Gruppioni2017).

The SMI/CAM 600 deg² shallow survey offers the potential to discover exceptionally luminous z > 5 galaxies for which SAFARI follow-up spectroscopy will be possible. Unlike submillimetre/millimetre surveys, in which high-redshift galaxies would often stand out in brightness, SMI mid-infrared surveys will see a large number of foreground galaxies, among which a small number of high-redshift objects will be hidden. Selecting z > 5 galaxies will therefore require a detailed analysis of good-quality multi-wavelength data covering the same fields like those mentioned in Section 4.5. Follow-up submillimetre/millimetre observations will especially be useful for discriminating z > 5 galaxies against lower-redshift galaxies.

It should also be noted that the SMI surveys have the potential to identify z > 5 HyLIRGs that may be missing in submillimetre/millimetre surveys. The all-sky mid-infrared survey with the Wide-Field Infrared Survey Explorer (WISE) has discovered hyper-luminous versions of dust-obscured galaxies (DOGs; Dey et al. Reference Dey2008), so-called Hot DOGs, up to a redshift of z = 4.59 (Eisenhardt et al. Reference Eisenhardt2012; Wu et al. Reference Wu2012; Tsai et al. Reference Tsai2015). These mid-infrared-selected Hot DOGs have bolometric luminosities of ≳ 10¹⁴ L_⊙, as luminous as the brightest submillimetre/millimetre-selected lensed infrared-luminous galaxies listed in Table 1. However, most of these galaxies do not appear to be lensed, suggesting that Hot DOGs are likely dominated by AGN (Tsai et al. Reference Tsai2015). Such AGN-dominated galaxies like Hot DOGs may constitute an important population of infrared-luminous galaxies at high redshift, which may be difficult to detect in the submillimetre/millimetres surveys unless they are also forming stars vigorously. The SMI surveys will be sensitive to this type of AGN-dominated galaxies and can detect them to higher redshift.