SLSN host galaxies throughout cosmic time

Cosmic evolution and metal aversion in super-luminous supernova host galaxies

S. Schulze, T. Krühler, G. Leloudas, J. Gorosabel, A. Mehner, J. Buchner, S. Kim, E. Ibar, R. Amorín, R. Herrero-Illana, J. P. Anderson, F. E. Bauer, L. Christensen, M. de Pasquale A. de Ugarte Postigo, A. Gallazzi, J. Hjorth, N. Morrell, D. Malesani, M. Sparre, B. Stalder, A. A. Stark, C. C. Thöne, J. C. Wheeler
Affiliations are listed at the end of the paper
E-mail: steve.schulze@weizmann.ac.il
Accepted 8 September 2017. Received 18 December 2016
Abstract

The SUperluminous Supernova Host galaxIES (SUSHIES) survey aims to provide strong new constraints on the progenitors of superluminous supernovae (SLSNe) by understanding the relationship to their host galaxies. We present the photometric properties of 53 H-poor and 16 H-rich SLSN host galaxies out to . We model their spectral energy distributions to derive physical properties, which we compare with other galaxy populations. At low redshift, H-poor SLSNe are preferentially found in very blue, low-mass galaxies with high average specific star-formation rates. As redshift increases, the host population follows the general evolution of star-forming galaxies towards more luminous galaxies. After accounting for secular evolution, we find evidence for differential evolution in galaxy mass, but not in the -band and the far UV luminosity ( confidence). Most remarkable is the scarcity of hosts with stellar masses above for both classes of SLSNe. In the case of H-poor SLSNe, we attribute this to a stifled production efficiency above solar metallicity. However, we argue that, in addition to low metallicity, a short-lived stellar population is also required to regulate the SLSN production. H-rich SLSNe are found in a very diverse population of star-forming galaxies. Still, the scarcity of massive hosts suggests a stifled production efficiency above solar metallicity. The large dispersion of the H-rich SLSNe host properties is in stark contrast to those of gamma-ray burst, regular core-collapse SN, and H-poor SLSNe host galaxies. We propose that multiple progenitor channels give rise to this sub-class.

keywords:
galaxies: evolution, mass function, starburst, star-formation, supernovae: general
pagerange: Cosmic evolution and metal aversion in super-luminous supernova host galaxiesDpubyear: XXX

1 Introduction

In the past decade, untargeted supernova (SN) surveys, e.g., the Texas SN Search (Quimby et al. 2005), the ROTSE SN Verification Project (Yuan et al. 2007), the Palomar Transient Factory (PTF; Law et al. 2009), and Pan-STARRS (PS; Tonry et al. 2012), discovered a new class of SNe with peak magnitudes exceeding  mag (Gal-Yam 2012). These so-called super-luminous supernovae have been a focus of SN science ever since, because of the opportunity they provide to study new explosion channels of very massive stars in the distant Universe (Howell et al. 2013; Cooke et al. 2012), the interstellar medium (ISM) in distant galaxies (Berger et al. 2012; Vreeswijk et al. 2014) and their potential use for cosmology (Inserra & Smartt 2014; Scovacricchi et al. 2016). In addition, SLSNe provide a new opportunity to pinpoint star-forming galaxies independently of galaxy properties, which can ultimately lead to a better understanding of galaxy evolution at the faint-end of luminosity and mass (Lunnan et al. 2014; Leloudas et al. 2015c; Angus et al. 2016; Chen et al. 2017; Perley et al. 2016b). Despite these prospects, SLSNe are very rare. At , one H-poor SLSN is expected to be produced for every 1000–20000 core-collapse SNe (hydrogen-rich SLSNe have a higher rate; Quimby et al. 2013a).

Phenomenologically, SLSNe can be classified by their hydrogen content into H-poor and H-rich SLSNe. The light curves of H-poor SLSNe (SLSNe-I), identified as a new class of transients by Quimby et al. (2011c), are  mag brighter and three-times broader than regular stripped-envelope SNe, but the shapes of their light-curves are similar (e.g., Quimby et al. 2011c; Inserra et al. 2013; Nicholl et al. 2015a). Early spectra of H-poor SLSNe show a characteristic w-shaped absorption feature at  Å  due to oxygen in the ejecta (Quimby et al. 2011c) that is usually not seen in Type Ibc SNe (e.g., Modjaz et al. 2009). About a month after maximum light, the ejecta cool down to temperatures typical of regular Type Ibc SNe at maximum light. At that point, SLSN spectra also exhibit absorption features similar to Type Ibc SNe (e.g., Pastorello et al. 2010; Inserra et al. 2013; Nicholl et al. 2014).

A subgroup of H-poor SLSNe shows exceptionally slowly-rising and slowly-declining light curves ( days,  days; Nicholl et al. 2015a), hereafter called slow-declining SLSN-I. In some cases the decay slope is comparable to that of the radioactive decay of Ni. Gal-Yam et al. (2009) argued that in the case of SN2007bi, the supernova was powered by the radioactive decay of several solar masses of Ni (Gal-Yam 2012), which were synthesised during a pair-instability SN (PISN) of a star with a zero-age-main-sequence (ZAMS) mass of (e.g., Fowler & Hoyle 1964; Barkat et al. 1967; Bisnovatyi-Kogan & Kazhdan 1967; Rakavy & Shaviv 1967; Fraley 1968; Heger et al. 2003; Woosley et al. 2007). However, the SN was discovered only shortly before it reached maximum light. Information about the rise time was not available, which is critical to distinguish between SN models. The well-sampled SLSNe PTF12dam and PS1-11ap, which were spectroscopically similar to SN2007bi at late times, had rise times that were incompatible with PISN models (Nicholl et al. 2013). This also cast doubt on the PISN interpretation of SN2007bi. However, recent findings by Kozyreva et al. (2017) showed that PISN models can predict short rise times similar to that of PTF12dam. Models of PISN spectra, on the other hand, are incompatible with the spectra of PTF12dam and SN2007bi (Dessart et al. 2013; Chatzopoulos et al. 2015; Jerkstrand et al. 2016).

The energy source powering H-poor SLSNe is highly debated. The most discussed models include magnetars formed during the collapse of massive stars (e.g., Kasen & Bildsten 2010; Inserra et al. 2013), the interaction of the SN ejecta with dense H-deficient circumstellar material (CSM) expelled by the progenitor prior to the explosion (Woosley et al. 2007; Blinnikov & Sorokina 2010; Chevalier & Irwin 2011; Chatzopoulos & Wheeler 2012; Quataert & Shiode 2012; Sorokina et al. 2016), PISNe, and pulsational PISNe (e.g., Woosley et al. 2007; Yan et al. 2015).

Hydrogen-rich SLSNe are characterised by an initial blue continuum and narrow Balmer lines, similar to classical Type IIn SNe (Schlegel 1990; Filippenko 1997; Kiewe et al. 2012) which are powered by the interaction of the supernova with its circumstellar material (e.g., Chevalier & Irwin 2011). Recent observations suggest a richer phenomenology. Spectra of the SNe 2008es and 2013hx showed broad H emission components and their light curves showed a linear decline after maximum, similar to normal IIL SNe (Gezari et al. 2009; Miller et al. 2009; Inserra et al. 2016). Another intriguing object is CSS121015:004244+132827 (hereafter called CSS121015). It firstly evolved as a H-poor SN but at 49 days after the maximum, its spectrum showed broad and narrow H emission lines (Benetti et al. 2014). These properties are different from superluminous type IIn SNe. Because of the similarities to Type II SNe, we label this subclass SLSN-II.

The possible diversity of SLSN progenitors suggests ZAMS masses up to a few hundred solar masses. Given the characteristic distance scale of SLSNe, a direct search for their progenitors is unfeasible. Alternatively, host observations have the potential to indirectly provide constraints on the progenitor population. The first systematic study of a sample of 17 H-poor and -rich SLSNe by Neill et al. (2011) suggested that the hosts are low-mass galaxies with high specific star-formation rates between and . However, these measurements are very uncertain because of the limited available wavelength coverage. This initial finding was supported by studies of the hosts of SN2010gx (Chen et al. 2013) and PS1-10bzj (Lunnan et al. 2013). Their spectroscopic observations also showed that both events occurred in low-metallicity galaxies with .

A survey of 31 H-poor SLSN host galaxies by Lunnan et al. (2014) consolidated the picture of H-poor SLSNe exploding in sub-luminous low-mass dwarf galaxies with median specific star-formation rates of . Furthermore, the preference for galaxies with a median metallicity of hinted at a stifled production efficiency at high metallicity (see also Leloudas et al. 2015c). Perley et al. (2016b) confirmed this trend by modelling the mass function of 18 SLSN-I hosts at from the PTF survey (see also Chen et al. 2017). Hubble Space Telescope observations of 16 hosts of H-poor SLSNe by Lunnan et al. (2015) revealed that the locations of H-poor SLSNe are correlated with the UV light distribution within their host galaxies. Yet, they are not as strongly clustered on the UV-brightest regions of their hosts than long-duration gamma-ray bursts (GRBs; see also Angus et al. 2016; Blanchard et al. 2016), which are also connected with the death of massive stars (e.g., Woosley 2012). Furthermore, on average, the interstellar medium of SLSN-I host galaxies is characterised by significantly weaker absorption lines than GRBs (Vreeswijk et al. 2014).

In 2012, we initiated the SUperluminous Supernova Host galaxIES (SUSHIES) survey (Leloudas et al. 2015c) to characterise a large set of host galaxies of H-poor and H-rich SLSNe over a large redshift range. The goals of this survey are to study SLSN host galaxies in context of other star-forming galaxies and to place constraints on the nature of their progenitors. To achieve this, our survey has spectroscopic and imaging components to characterise the integrated host properties, such as mass, metallicity, star-formation rate, age of the stellar populations and dust attenuation.

In the first SUSHIES sample paper, Leloudas et al. (2015c) discussed the spectroscopic properties of 17 H-poor and 8 H-rich SLSN host galaxies. We showed that the host galaxies of H-poor SLSNe are characterised by hard ionisation fields, low metallicity and very high specific star-formation rates. A high number () of H-poor SLSNe at occurred in extreme emission-line galaxies (e.g., Atek et al. 2011; Amorín et al. 2014, 2015), which represent a short-lived phase in galaxy evolution following an intense starburst. Moreover, in Thöne et al. (2015) we performed spatially resolved spectroscopy of the host of PTF12dam, the most extreme host galaxy in the sample with high signal to noise, and found strong evidence for a very young stellar population at the explosion site with an age of  Myr. These findings let us conclude in Leloudas et al. (2015c) that the progenitors of SLSNs are possibly the very first stars to explode in a starburst, at an earlier evolutionary stage than GRB progenitors. Therefore, not only metallicity but also age is likely a critical condition for the production of SLSN progenitors. Chen et al. (2017) and Perley et al. (2016b) questioned the importance of the age and proposed that metallicity is the primary factor for SLSN-I progenitors.

While H-poor SLSNe are preferentially found in rather extreme environments, the findings by Leloudas et al. (2015c) and Perley et al. (2016b) point to a weaker dependence on environment properties for H-rich SLSNe, e.g., higher average metallicities and softer ionisation states.

In this second sample paper of the SUSHIES survey, we present photometric data of a sample of 53 H-poor and 16 H-rich SLSN host galaxies out to , including almost every SLSN reported in the literature and detected before 2015. The scope of this paper is to provide distribution functions of physical properties, such as luminosities, masses of the stellar populations and star-formation rates, to investigate their redshift evolution and to compare these results to other samples of starburst galaxies.

Throughout the paper, we adopt a flat CDM cosmology with , , (Planck Collaboration 2014). Uncertainties and dispersions are quoted at confidence. We refer to the solar abundance compiled in Asplund et al. (2009).

Object R. A. Dec. Redshift Type Decline time Reference
(J2000) (J2000) (mag) scale  (days)
Spectroscopic sample (23)
PS1-10bzj 03:31:39.83 27:47:42.2 0.649 SLSN-I 0.01 37.3(fast) [1, 2]
PS1-11ap 10:48:27.73 57:09:09.2 0.524 SLSN-I 0.01 87.9 (slow) [2, 3]
PTF09cnd 16:12:08.94 51:29:16.1 0.258 SLSN-I 0.02 75.3 (slow) [2, 4]
PTF10heh 12:48:52.04 13:26:24.5 0.338 SLSN-IIn 0.02 [5]
PTF10hgi 16:37:47.04 06:12:32.3 0.099 SLSN-I 0.07 35.6 (fast) [2, 6, 7]
PTF10qaf 23:35:42.89 10:46:32.9 0.284 SLSN-IIn 0.07 [8]
PTF10vqv 03:03:06.84 01:32:34.9 0.452 SLSN-I 0.06 [9]
PTF11dsf 16:11:33.55 40:18:03.5 0.385 SLSN-IIn 0.01 [10]
PTF12dam 14:24:46.20 46:13:48.3 0.107 SLSN-I 0.01 72.5 (slow) [2, 11]
SN1999as 09:16:30.86 13:39:02.2 0.127 SLSN-I 0.03 [8, 12]
SN1999bd 09:30:29.17 16:26:07.8 0.151 SLSN-IIn 0.03 [8, 13]
SN2006oz 22:08:53.56 00:53:50.4 0.396 SLSN-I 0.04 [14]
SN2006tf11footnotemark: 12:46:15.82 11:25:56.3 0.074 SLSN-IIn 0.02 [15]
SN2007bi22footnotemark: 13:19:20.00 08:55:44.0 0.128 SLSN-I 0.02 84.5 (slow) [2, 16, 17]
SN2008am 12:28:36.25 15:35:49.1 0.233 SLSN-IIn 0.02 [18]
SN2009jh33footnotemark: 14:49:10.08 29:25:11.4 0.349 SLSN-I 0.01 60.6 (slow) [2, 4]
SN2010gx44footnotemark: 11:25:46.71 08:49:41.4 0.230 SLSN-I 0.03 29.1 (fast) [2, 4, 19]
SN2010kd 12:08:01.11 49:13:31.1 0.101 SLSN-I 0.03 [20, 21]
SN2011ke55footnotemark: 13:50:57.77 26:16:42.8 0.143 SLSN-I 0.01 25.7 (fast) [2, 6]
SN2011kf66footnotemark: 14:36:57.53 16:30:56.6 0.245 SLSN-I 0.02 28.5 (fast) [2, 6]
SN2012il77footnotemark: 09:46:12.91 19:50:28.7 0.175 SLSN-I 0.02 23.2 (fast) [2, 6]
SNLS06D4eu 22:15:54.29 18:10:45.6 1.588 SLSN-I 0.02 [22]
SSS12081088footnotemark: 23:18:01.82 56:09:25.7 0.156 SLSN-I 0.02 30.2 (fast) [2, 23]
Non-spectroscopic sample (46)
CSS10021799footnotemark: 10:29:12.56 40:42:20.0 0.147 SLSN-IIn 0.01 [24]
CSS1210151010footnotemark: 00:42:44.34 13:28:26.5 0.286 SLSN-II 0.07 37.8 (fast) [2, 25]
CSS1409251111footnotemark: 00:58:54.11 18:13:22.2 0.460 SLSN-I 0.06 [26]
DES14S2qri 02:43:32.14 01:07:34.2 1.500 SLSN-I 0.03 [27]
DES14X2byo 02:23:46.93 06:08:12.3 0.869 SLSN-I 0.03 [28]
DES14X3taz 02:28:04.46 04:05:12.7 0.608 SLSN-I 0.02 [29]
iPTF13ajg 16:39:03.95 37:01:38.4 0.740 SLSN-I 0.01 62.0 (slow) [2, 30]
LSQ12dlf1212footnotemark: 01:50:29.80 21:48:45.4 0.255 SLSN-I 0.01 35.4 (fast) [2, 23]
LSQ14an 12:53:47.83 29:31:27.2 0.163 SLSN-I 0.07 [31]
LSQ14mo 10:22:41.53 16:55:14.4 0.2561 SLSN-I 0.06 27.3 (fast) [2, 32]
LSQ14bdq 10:01:41.60 12:22:13.4 0.345 SLSN-I 0.06 71.2 (slow) [2, 33]
LSQ14fxj 02:39:12.61 03:19:29.6 0.360 SLSN-I 0.03 [34]
MLS1211041313footnotemark: 02:16:42.51 20:40:08.5 0.303 SLSN-I 0.15 [35, 36]
PS1-10ky 22:13:37.85 01:14:23.6 0.956 SLSN-I 0.03 32.5 (fast) [2, 37]
PS1-10pm 12:12:42.20 46:59:29.5 1.206 SLSN-I 0.02 [38]
PS1-10ahf 23:32:28.30 00:21:43.6 1.100 SLSN-I 0.03 [38]
PS1-10awh 22:14:29.83 00:04:03.6 0.909 SLSN-I 0.07 [37]
PS1-11tt 16:12:45.78 54:04:17.0 1.283 SLSN-I 0.01 [39]
PS1-11afv 12:15:37.77 48:10:48.6 1.407 SLSN-I 0.01 [39]
PS1-11aib 22:18:12.22 01:33:32.0 0.997 SLSN-I 0.04 [39]
PS1-11bam 08:41:14.19 44:01:57.0 1.565 SLSN-I 0.02 [40]
PS1-11bdn 02:25:46.29 05:06:56.6 0.738 SLSN-I 0.02 [39]
PS1-12zn 09:59:49.62 02:51:31.9 0.674 SLSN-I 0.02 [39]
PS1-12bmy 03:34:13.12 26:31:17.2 1.566 SLSN-I 0.01 [39]
PS1-12bqf 02:24:54.62 04:50:22.7 0.522 SLSN-I 0.02 [39]
PS1-13gt 12:18:02.03 47:34:46.0 0.884 SLSN-I 0.02 [39]
PTF09atu 16:30:24.55 23:38:25.0 0.501 SLSN-I 0.04 [4]
PTF11rks 01:39:45.51 29:55:27.0 0.190 SLSN-I 0.04 22.3 (fast) [2, 6, 41]
SCP06F6 14:32:27.40 33:32:24.8 1.189 SLSN-I 0.01 39.8 (fast) [2, 42]
SN2003ma 05:31:01.88 70:04:15.9 0.289 SLSN-IIn 0.31 [43]
SN2005ap 13:01:14.83 27:43:32.3 0.283 SLSN-I 0.01 28.8 (fast) [2, 44]
SN2006gy 03:17:27.06 41:24:19.5 0.019 SLSN-IIn 0.14 [45]
SN2007bw1414footnotemark: 17:11:01.99 24:30:36.4 0.140 SLSN-IIn 0.04 [46]
SN2008es1515footnotemark: 11:56:49.13 54:27:25.7 0.205 SLSN-II 0.01 38.0 (fast) [2, 47, 48]
SN2008fz1616footnotemark: 23:16:16.60 11:42:47.5 0.133 SLSN-IIn 0.04 [49]
SN2009de1717footnotemark: 13:00:37.49 17:50:57.0 0.311 SLSN-I 0.04 [50, 51, 52]
SN2009nm1818footnotemark: 10:05:24.54 51:16:38.7 0.210 SLSN-IIn 0.01 [53, 54]
Table 1: Properties of the super-luminous supernovae in our sample

Properties of the super-luminous supernovae in our sample Object R. A. Dec. Redshift Type Decline time Reference (J2000) (J2000) (mag)  (days) SN2011cp1919footnotemark: 07:52:32.61 21:53:29.7 0.380 SLSN-IIn 0.05 [55] SN2011ep2020footnotemark: 17:03:41.78 32:45:52.6 0.280 SLSN-I 0.02 [56] SN2013dg2121footnotemark: 13:18:41.38 07:04:43.1 0.265 SLSN-I 0.04 30.7 (fast) [2, 23] SN2013hx2222footnotemark: 01:35:32.83 57:57:50.6 0.130 SLSN-II 0.02 33.6 (fast) [2, 57] SN2013hy2323footnotemark: 02:42:32.82 01:21:30.1 0.663 SLSN-I 0.03 [58] SN2015bn2424footnotemark: 11:33:41.57 00:43:32.2 0.110 SLSN-I 0.02 [59] SN1000+0216 10:00:05.87 02:16:23.6 3.899 SLSN-I 0.02 [60] SN2213-1745 22:13.39.97 17:45:24.5 2.046 SLSN-I 0.02 [60] SNLS07D2bv 10:00:06.62 02:38:35.8 SLSN-I 0.02 [22] Note. – The coordinates refer to the positions of the supernovae. The Galactic extinction measurements are taken from Schlafly & Finkbeiner (2011). We divide the sample into the spectroscopic sample (23 objects) presented in Leloudas et al. (2015c) and in a non-spectroscopic sample (46 objects). The decay-time scale is defined as the time when the luminosity of the pseudo-bolometric light curve dropped to . We divide the sample into fast and slow decliners if and days, respectively.
The classifications of SN1000+0213 and SN2213-1745 are based on photometry. The light curve of SN1000+0213 shows a bump before the main emission similar to H-poor SLSNe SN2006oz and LSQ14bdq (for details see Leloudas et al. 2012; Nicholl et al. 2015a).
Alternative SN names:  CSS070320:124616+112555;  SNF20070406-008;  CSS090802:144910+292510, PTF09cwl;  CSS100313:112547-084941, PTF10cwr;  CSS110406:135058+261642, PTF11dij, PS1-11xk;  CSS111230:143658+163057;  CSS120121:094613+195028, PS1-12fo;  SSS120810:231802-560926;  CSS100217:102913+404220;  CSS121015:004244+132827;  CSS140925:005854+181322;  SSS120907:015030-214847;  MLS121104:021643+204009, LSQ12fzb;  SNF20070418-020;  ROTSE3 J115649.1+542725;  CSS080922:231617+114248;  CSS090102:130037+175057, PSN K0901-1;  CSS091120:100525+511639;  MLS110426:075233+215330, PSN J07523261+2153297;  CSS110414:170342+324553;  CSS130530:131841-070443, MLS130517:131841-070443;  SMT J013533283-5757506;  DES13S2cmm;  CSS141223:113342+004332, MLS150211:113342+004333, PS15ae References. – [1]: Lunnan et al. (2013); [2]: Nicholl et al. (2015a); [3]: McCrum et al. (2014); [4]: Quimby et al. (2011c); [5]: Quimby et al. (2010a); [6]: Inserra et al. (2013); [7]: Leloudas et al. (2015c); [8]: Gal-Yam (2012); [9]: Quimby et al. (2010b); [10]: Quimby et al. (2011a); [11]: Nicholl et al. (2013); [12]: Knop et al. (1999); [13]: Nugent et al. (1999); [14]: Leloudas et al. (2012); [15]: Smith et al. (2008); [16]: Gal-Yam et al. (2009); [17]: Young et al. (2010); [18]: Chatzopoulos et al. (2011); [19]: Pastorello et al. (2010); [20]: Vinko et al. (2012); [21]: Quimby et al. (2013b); [22]: Howell et al. (2013); [23]: Nicholl et al. (2014); [24]: Drake et al. (2011a); [25]: Benetti et al. (2014); [26]: Campbell et al. (2014); [27]: Castander et al. (2015); [28]: Graham et al. (2014); [29]: Smith et al. (2016); [30]: Vreeswijk et al. (2014); [31]: Leget et al. (2014); [32]: Leloudas et al. (2015b); [33]: Nicholl et al. (2015b); [34]: Smith et al. (2014); [35]: Drake et al. (2012); [36]: Fatkhullin & Gabdeev (2012); [37]: Chomiuk et al. (2011); [38]: McCrum et al. (2015); [39]: Lunnan et al. (2014); [40]: Berger et al. (2012); [41]: Quimby et al. (2011b); [42]: Barbary et al. (2009); [43]: Rest et al. (2011); [44]: Quimby et al. (2007); [45]: Smith et al. (2007); [46]: Agnoletto (2010); [47]: Gezari et al. (2009); [48]: Miller et al. (2009); [49]: Drake et al. (2010); [50]: Drake et al. (2009b); [51]: Drake et al. (2009a); [52]: Moskvitin et al. (2010); [53]: Drake et al. (2009c); [54]: Christensen et al. (2009); [55]: Drake et al. (2011b); [56]: Graham et al. (2011a); [57]: Inserra et al. (2016); [58]: Papadopoulos et al. (2015); [59]: Nicholl et al. (2016) [60]: Cooke et al. (2012);

Table 2: continued

2 Sample definition, observations and data reduction

2.1 Sample definition

Among all SLSNe reported in the literature (), we selected those that were discovered before the end of 2014 and announced before April 2015. Therefore, many of the SLSNe published recently by Perley et al. (2016b) are not included in this paper. In addition, we screened the Asiago Supernova catalogue (Barbon et al. 2010) for objects with an absolute magnitude of significantly brighter than  mag and spectroscopic information. This revealed two additional H-poor SLSNe, SNe 2009de and 2011ep (Drake et al. 2009b; Moskvitin et al. 2010; Graham et al. 2011a), and two H-rich SLSNe, SNe 2009nm and SN2011cp (Drake et al. 2009c; Christensen et al. 2009; Drake et al. 2011c, d; Graham et al. 2011b). The SN properties are summarised in Table 1.

Our final sample comprises of 53 H-poor and 16 H-rich SLSNe. The H-poor sample includes 7 slow-declining H-poor SLSNe, while the H-rich sample includes the SLSNe-II CSS121015, SN2008es and SN2013hx. The size of the final sample is not only a factor of larger than the SLSN host sample presented in Perley et al. (2016b) but includes a large population of hosts at (which is the highest redshift in Perley et al. 2016b). Figure 1 displays the redshift distribution of our sample. It covers a redshift interval from to with a singular object at (SN1000+0216; Cooke et al. 2012). The redshift distribution of the H-poor sample covers the full range and has a median of . The H-rich sample only extends to and has a median of .

2.2 Observations

A fundamental goal of our survey is to secure multi-band data from the rest-frame UV to NIR, to model the spectral energy distributions of the host galaxies. To ensure a sufficient wavelength coverage and data quality, we aimed to have at least one observation of the rest-frame UV and of the NIR and two observations of the rest-frame optical, if a galaxy was brighter than  mag.

To optimise the observing campaign, we queried the VizieR database (Ochsenbein et al. 2000) and public archives for available catalogues and data, such as the ESO, Gemini and Subaru archives. Our primary source catalogues are from the Canada-France-Hawaii Telescope Legacy Survey (CFHTLS; Hudelot et al. 2012), the Cosmological Evolution Survey (COSMOS; Scoville et al. 2007), the Galaxy Evolution Explorer (GALEX; Martin et al. 2005), the Sloan Digital sky survey (SDSS; York et al. 2000), the UKIRT Infrared Deep Sky Survey (UKIDSS; Lawrence et al. 2007) and the Wide-field Infrared Survey Explorer (WISE; Wright et al. 2010).111We included WISE data of only a few hosts. These catalogues were complemented by the Coma Cluster catalogue (Adami et al. 2006), the UltraVISTA catalogue (McCracken et al. 2012), the VISTA Deep Extragalactic Observations survey (VIDEO; Jarvis et al. 2013) and the VIRMOS deep imaging survey (VIRMOS; Le Fèvre et al. 2004). Furthermore, we incorporated measurements previously reported in Inserra et al. (2013), Lunnan et al. (2014), Nicholl et al. (2014), Vreeswijk et al. (2014) and Angus et al. (2016).

Between 2012 and 2016, we used observing proposals at the 6.5-m Magellan/Baade Telescope (PI: Schulze, Kim),222Programme IDs: CN2013A-195, CN2013B-70, CN2014A-114, CN2014B-127, CN2014B-102, CN-2015A-129, CN2015A-143, CN-2015B-87, CN2015B-99, CN2016A-108, and CN2016B-98 ESO’s 8.2-m Very Large Telescope (VLT; PI: Leloudas, Krühler),333Programme IDs: 089.D-0902, 091.A-0703, 091.D-0734, and 290.D-5139 the 10.4-m GTC and 3.5-m CAHA telescope (PI: Gorosabel) and the 0.3-m UV/Optical Telescope (UVOT; Roming et al. 2005) onboard the Swift satellite (Gehrels et al. 2004, PI: Leloudas) to obtain rest-frame UV, optical and NIR data. In the subsequent sections, we briefly summarise each campaign.

Our Magellan campaign was performed between 2012 and 2016 with the 6.5-m Baade telescope equipped with the optical wide-field Inamori-Magellan Areal Camera and Spectrograph (IMACS; Dressler et al. 2011), the Parallel Imager for Southern Cosmological Observations (PISCO; Stalder et al. 2014), and the near-infrared (NIR) camera FourStar (Persson et al. 2013). The optical data were secured in , primarily with the IMACS f/2 camera, but also with the IMACS f/4 camera and PISCO. The near infrared observations were performed in and .

The ESO VLT observations were taken in visitor and service mode. The visitor run took place between 29 May and 2 June 2013. We used the FOcal Reducer and Spectrograph 2 instrument (FORS2; Appenzeller et al. 1998), equipped with the red-sensitive CCD to secure data in . In addition, we obtained and band imaging with the High Acuity Wide field K-band Imager (HAWK-I; Pirard et al. 2004; Casali et al. 2006; Kissler-Patig et al. 2008). Additional optical and NIR data were obtained with FORS2, the Infrared Spectrometer And Array Camera (ISAAC; Moorwood et al. 1998) and HAWK-I in queue mode.

The CAHA and GTC campaigns primarily focused on targets on the northern hemisphere. The CAHA observing programme was carried out with the 4-channel Bonn University Simultaneous CAmera (BUSCA; Reif et al. 1999) in at the 3.5-m CAHA telescope in 2012. We also used the infrared wide-field camera Omega2000 (Kovács et al. 2004) to secure and band observations between 2013 and 2015 and also in and band for a few targets. The objective of the campaign at the 10.4-m GTC telescope was to secure deep imaging of SNe 2008es and 2009jh with the Optical System for Imaging and low-Intermediate-Resolution Integrated Spectroscopy (OSIRIS; Cepa et al. 2000) camera.

Rest-frame UV data are critical to break degeneracies in the SED modelling. For objects at , observations in or bluer filters are needed to probe the UV. GALEX provided critical rest-frame UV data for most objects. In addition, we secured UV photometry of five fields with the UV/optical telescope UVOT on board the Swift satellite in 2014 and incorporated archival UVOT data of a further SLSN.

These core observing campaigns were complemented by smaller observing programmes that targeted selected host galaxies. We observed the field of SN2005ap with the Andalucia Faint Object Spectrograph and Camera (ALFOSC) at the 2.54-m Nordic Optical Telescope and the field of SN2007bi with ALFOSC and the 7-channel imager Gamma-Ray Burst Optical/Near-Infrared Detector (GROND; Greiner et al. 2008) at the 2.2-m Max-Planck-Gesellschaft telescope.

To place limits on the total star-formation rate, we used 1.4 GHz data from the VLA Faint Images of the Radio Sky at Twenty-Centimeters survey (FIRST; Becker et al. 1995), the NRAO VLA Sky Survey (NVSS,  GHz; Condon et al. 1998), and 843 MHz data from the Sydney University Molonglo Sky Survey (SUMSS; Bock et al. 1999). In addition, we secured continuum observations of MLS121104, SN2005ap and SN2008fz with the Karl Jansky Very Large Array (JVLA; PI: Ibar).444Programme ID: 15A-224 The continuum observations were performed in L band in the most extended A-configuration in July and September 2015. The frequency was centred at 1.5 GHz with a total synthesised bandwidth of 1 GHz. We used the standard flux and bandwidth calibrator 3C48 for all the sources except SN2005ap, for which we used 3C286 instead. For phase calibration purposes we used bright nearby point-like sources from the VLA calibrator list (MLS121104: J0238+1636; SN2005ap: J1310+3220; and SN2008fz: J2330+1100). The key properties of each observation is reported in Tables 9.

2.3 Data reduction

Figure 1: The redshift distribution of the SUSHIES survey. For 21 H-poor SLSNe, information about the decline time-scale are available. The region hatched by ‘//’ displays the redshift distribution of the fast-decliners and the region highlighted by ‘’ signifies the redshift distribution of the slow-decliners. The redshift distribution of the three SLSNe-II, CSS121015, SN2008es and SN2013hx, are highlighted by ‘o’. The median redshifts of the H-poor and H-rich sample are (solid vertical line) and (dashed vertical line), respectively.

We reduced all data in a consistent way with standard routines in IRAF (Tody 1986). The typical steps are i) bias/overscan subtraction, ii) flat-fielding, iii) fringe correction, iv) stacking of individual images and v) astrometric calibration. For a few instruments we used instrument specific software packages: the GEMINI IRAF package, the GROND pipeline (Yoldaş et al. 2008; Krühler et al. 2008), PHOTPIPE for PISCO data (Bleem et al. 2015), SDFRED1 and SDFRED2 for Subaru Suprime-Cam data (Yagi et al. 2002; Ouchi et al. 2004), THELI version 2.10.0 (Erben et al. 2005; Schirmer 2013) for the FourStar data, VLT instrument pipelines for HAWK-I (version 1.8.18) and ISAAC (version 6.1.3) data,555http://www.eso.org/sci/software/cpl/esorex.html and a customised pipeline for the Magellan/IMACS data. The world-coordinate systems were calibrated with astrometry.net version 0.5 (Lang et al. 2010).

UVOT data were retrieved from the Swift Data Archive.666http://www.swift.ac.uk/swift_portal/ We used the standard UVOT data analysis software distributed with HEAsoft version 6.12, along with the standard calibration data.777http://heasarc.nasa.gov/lheasoft/

The JVLA data were reduced using the Common Astronomy Software Applications package (CASA; McMullin et al. 2007) and consisted of careful data flagging and standard flux, bandwidth and phase calibration. No self-calibration was performed to the data. The obtained flux density root mean squares (r.m.s.) of the images are summarised in Table 23.

3 Methods

3.1 Host identification

We aligned our host-only images with the original SN images that we retrieved from archives with Gaia version 4.4.6.888http://starlink.eao.hawaii.edu/starlink/2015ADownload The average alignment accuracy was . We neither found (suitable) public data for 13 SNe from PanSTARSS, nor for SNe 2006tf, 2009de, 2009nm and 2011cp (in total 17/69 objects). For those objects we relied on the reported SN positions. Although this added an uncertainty to the host identification, the SN positions always coincided with a galaxy, which we assume is the host galaxy.

3.2 Photometry

We developed a Python programme that is based on Source Extractor version 2.19.5 (Bertin & Arnouts 1996) to perform seeing matched aperture photometry. To measure the total flux of the given object, the source radius was typically 2–4 times the full-width at half maximum (FWHM) of the stellar PSF. In case another object was close to the SN position or if the host had a large angular diameter, we adjusted the extraction radius accordingly. If a host evaded detection in all bands, we measured the flux and its uncertainty at the SN position using an aperture with a radius of . Those measurements have very large uncertainties but they can be easily included in the SED modelling in contrast to upper limits.

Once an instrumental magnitude was established, it was photometrically calibrated against the brightness of several standard stars measured in a similar manner or tied to the SDSS DR8 (Aihara et al. 2011) and the AAVSO (American Association of Variable Star Observers) Photometric All-Sky Survey (APASS) DR9 (Henden et al. 2016) catalogues. For Bessell/Johnson/Cousins filters, we converted the photometry of stars in the SDSS catalogue from SDSS using the Lupton colour equations.999http://www.sdss.org/dr5/algorithms/sdssUBVRITransform.html In the NIR (), the photometry was tied to 2MASS. The UVOT photometry was performed with the programme uvotsource. UVOT zeropoints are defined for an aperture with a diameter of . We translated these zeropoints into those of our requested apertures by applying simple aperture correction methods for stars.

Finally, the measurements were corrected for Galactic extinction using the extinction maps by Schlafly & Finkbeiner (2011) and transformed into the AB system using Blanton & Roweis (2007) and Breeveld et al. (2011).

In total, we measured the brightness (and limits for the non-detections) of 53 of the 69 objects, which also includes the re-evaluation of 27 individual data sets from the Two Micron All Sky Survey (2MASS), CFHTLS and SDSS, as well as several archival data sets. In addition, we augmented the photometry of 31 objects by literature values, such as GALEX, Pan-STARRS and WISE data. Owing to GALEX’s and WISE’s large point-spread functions, we only included their photometry if a contamination by neighbouring objects could be excluded. Among the 16 objects whose photometry is entirely based on literature results, four galaxies are in the footprint of the COSMOS survey: PS1-12zn, PS1-12bqf, SN1000+0213 and SNLS07D2bv. Their photometry is discussed here for the first time. Table 9 summarises the photometry of each object.

3.3 Spectral-energy distribution fitting

We modelled the SEDs with Le Phare (Arnouts et al. 1999; Ilbert et al. 2006),101010http://www.cfht.hawaii.edu/~arnouts/LEPHARE using a grid of galaxy templates based on Bruzual & Charlot (2003) stellar population-synthesis models with a Chabrier IMF (Chabrier 2003). The star-formation history was approximated by a declining exponential function of the form , where is the age of the stellar population and the e-folding time-scale of the star-formation history (varied in eight steps between 0.1 and 15 Gyr). Furthermore, we assumed the Calzetti dust attenuation curve (Calzetti et al. 2000). For a description of the galaxy templates, physical parameters of the galaxy fitting, and their error estimation, we refer to Krühler et al. (2011).111111The templates used in this paper do not account for possible binary star evolution, which could substantially alter SEDs (more hard UV photons; e.g., Stanway et al. 2016).

As an extension to Krühler et al. (2011), we relaxed the analysis threshold of the galaxy mass to (which is pushing the definition of a galaxy), because previous studies showed that SLSNe can occur in very low-mass galaxies (Lunnan et al. 2014; Leloudas et al. 2015c; Angus et al. 2016). We modified the gas component in Le Phare by incorporating the observed relationship between line flux and SFR for [O ii] and [O iii] by Krühler et al. (2015). The attenuation of the ionised gas component was linked to the stellar attenuation via by Calzetti et al. (2000). All attenuation measurements are reported for . Finally, we used the high-resolution BC03 templates, which are defined over 6900 wavelength points instead of 1221 wavelength points from to 160 m. To account for zeropoint offsets in the cross-calibration and absolute flux scale, we added a systematic error of 0.05 mag in quadrature to the uncertainty introduced by photon noise. For GALEX, UVOT and -band data this systematic error was increased to 0.1 mag.

The absolute magnitudes were computed directly by convolving the filter response functions with the best-fit template. To compute the corresponding error in the rest-frame bandpass , we interpolated between the errors of the apparent magnitudes and of the observed band-pass and , respectively, via:

where is the central wavelength of the observer-frame bandpass and in the rest-frame of the SLSNe. In case a rest-frame bandpass lies blueward/redward of the observation in the bluest/reddest filter, we set the error to the error of the observation in the bluest/reddest filter.

Our observations were characterised by a large set of different filters, of which several have similar bandpasses. To simplify the fitting, we homogenised the filter set. Specifically, we set the filter response function of , , , to , , , , , to , , , to , , , , to ’, , , , , to , to , to , to , , to , to , to , to Spitzer/, and to Spitzer/. It can be seen from our fits (Figs. 2, 18 and 22), and quality of the derived host properties (Table 5), that the impact of these assumptions is negligible.

Studies of SLSN host galaxies and extreme emission-line galaxies (e.g., Amorín et al. 2015) showed that emission lines can significantly affect the SED fitting. To quantify this effect, we repeated the SED fitting for our spectroscopic sample (Leloudas et al. 2015c; Table 1). The contribution of the emission line on the photometry in filter is given by

where is the flux density of the emission line , is the flux density of the stellar continuum and is the transmission function of the filter . The strength of an emission line can be characterised by its equivalent width, EW, hence . Assuming that all emission lines are narrow compared to the width of the broad-band filter, the above expression simplifies to

where is the filter response function of filter at the wavelength of the emission line (in the air reference frame) and is the effective width of the filter. In contrast to the SED fitting, it was necessary to use the exact filter transmission function of each instrument.

We subtracted the contribution of H–H, [O ii], [O iii], [N ii], [Ne ii] and [S ii] from the measured brightness in the broadband filter. Afterwards, we explicitly switched off the contribution from the ionised gas of H ii regions in Le Phare and repeated the fits with the emission-line-subtracted SEDs. The result of this experiment is discussed in Sect. 4.1.2.

3.4 Ensemble statistics

To compare observed distributions with distributions of other galaxy samples (parent distributions), such as extreme emission-line galaxies (hereafter EELGs), GRBs and SNe, we performed an Monte-Carlo (MC) simulation as follows. Each SLSN host measurement was represented by a normal distribution centred at the observed value and with a width () determined from the asymmetric error or a uniform distribution between the upper limit and the smallest/faintest value in the sample for those objects with upper limits only. A two-sided Anderson-Darling (AD) test was performed between the resampled distributions and the parent distributions, using the R package kSamples. This process was repeated 10 000 times and a mean AD value obtained. We rejected the null hypothesis of two distributions being drawn from the same parent distribution if the corresponding chance probability was smaller than 0.01.

To complement the one-dimensional Anderson-Darling tests, we also performed two-dimensional tests in the mass-SFR plane. We first computed the mean mass and SFR of the SLSN-host sample. After that, we bootstrapped 10 000 samples of size from the other galaxy samples, where is the number of SLSNe in the given redshift interval, and computed the mean mass and SFR of each bootstrapped sample. Measurement errors were propagated through a MC simulation as described above. Finally, we computed the region that contained 99% of all realisations using the python package corner.py (Foreman-Mackey 2016). If the estimator of the SLSN sample did not fall in that region, the chance probability is less than 0.01 and we rejected the null hypothesis of both distributions being statistically similar.

For each statistical test, we also performed a two-sided AD test on the redshift distributions to minimise systematic errors introduced by cosmic evolution, similar to Japelj et al. (2016).

We extract robust estimates of the ensemble distribution functions with a Bayesian approach, which incorporates the varying and asymmetric measurement uncertainties of individual sources and the limited sample size. For this we fit to the sample measurements in a quantity (e.g., in or SFR) a normal distribution. We constrain its parameters, the mean and standard deviation , with a likelihood defined as the product of convolutions of that distribution and the measurement probability distributions. The fit uncertainties were obtained with the MultiNest package (Feroz et al. 2013) through the python package PyMultiNest (Buchner et al. 2014). Flat priors were assumed on and .

3.5 Comparison samples

We built several comparison samples to put SLSN host galaxies in context with the cosmic star-formation history and to better understand the peculiar conditions that gave rise to this class of stellar explosion.

Core-collapse supernova host galaxies:

Because of the connection between SLSNe and massive stars, we compiled core-collapse supernova (CCSN) host galaxy samples. As in Leloudas et al. (2015c), we used SNe from untargeted (with respect to galaxies) surveys. At , we use objects studied in Leloudas et al. (2011), Sanders et al. (2012) and Stoll et al. (2013). All SNe in these samples have robust spectroscopic classifications. The combined sample consists of 44 type Ib/c SNe and 46 type II SNe. These studies provide multi-band data, which are primarily based on SDSS photometry and also spectroscopy for a number of hosts. We adopt the SED modelling by Leloudas et al. (2015c) for the Leloudas et al. (2011) and Sanders et al. (2012) samples. Note, the spectral energy distributions in Stoll et al. (2013) were modelled with the FAST stellar population synthesis code (Kriek et al. 2009) with the Bruzual & Charlot (2003) templates and a Salpeter IMF. We reduced their SFRs and galaxy masses by a factor of 1.8, to convert from a Salpeter to a Chabrier IMF, used in this paper (Kennicutt 1998).

To expand the SN sample to redshifts larger than , where most of our SLSNe are found, we added the SN sample from the Great Observatories Origins Deep Survey (GOODS) and Probing Acceleration Now with Supernovae (PANS) surveys (Riess et al. 2004). GOODS/PANS were HST surveys to detect Type Ia SNe at high redshift. This survey also located 58 distant CCSNe between and (the median being ). In contrast to the low- samples, their classification relied on photometric data. The method allowed a distinction between Type Ia and CCSNe, but not a categorisation into sub-types. Thanks to the overlap with the GOODS field, each SN host has deep rest-frame UV to NIR data. We adopt the results of the SED modelling by Svensson et al. (2010). Note, these authors modelled the SEDs with their own software that uses observed SEDs of local galaxies and SEDs produced with various spectral synthesis codes as templates. Furthermore, they assumed a Salpeter IMF. Similar to Stoll et al. (2013), the SFRs and the masses were reduced by a factor of 1.8 to convert from a Salpeter to a Chabrier IMF.

GRB host galaxies:

A member of our team (T. Krühler) collected multi-band data of long GRBs. These GRBs are selected to be part of one of the following complete GRB samples: GROND 4-hour sample (Greiner et al. 2011), TOUGH survey (The Optically Unbiased GRB Host Galaxy survey; Hjorth et al. 2012), BAT-6 (Salvaterra et al. 2012) or SHOALS (Swift Gamma-Ray Burst Host Galaxy Legacy Survey; Perley et al. 2016b). The individual measurements are reported in Krühler & Schady (2017). Among all hosts, we selected those at (52 in total). At these redshifts, it is relatively easy to secure the GRB redshift, because of the sparsity of dust-obscured bursts at , and to build host samples with a high detection completeness. The SEDs of this sample were analysed in a similar way as our SLSN host galaxy sample.

Sample Selection criteria Number of Redshift Which properties
objects interval used?
Core-collapse supernova host galaxies (total number 265)
Leloudas et al. (2011) Ib/c SNe, detected by untargeted surveys 12 , mass, SFR11footnotemark:
(L11) spectroscopic classification
Sanders et al. (2012) Ib/c SNe, detected by untargeted surveys 31 , mass, SFR11footnotemark:
(S12) spectroscopic classification
Svensson et al. (2010) GOODS SN sample 165 , mass, SFR
photometric SN classification
Stoll et al. (2013) first-year PTF CCSN sample 58 , mass, SFR
(S13) primarily Type II SNe
Extreme emission-line galaxies (total number 227)
Amorín et al. (2014) VUDS survey (Le Fèvre et al. 2015), 31 colour, , ,
 mag mass, SFR
Amorín et al. (2015) zCOSMOS survey,  mag 165 colour, , ,
mass, SFR
Atek et al. (2011) WISPS survey (Atek et al. 2010), 9 mass, SFR
Maseda et al. (2014) 3D-HST survey (Brammer et al. 2012), colour selection 22 mass, SFR
emission-lines do not fall in the NIR band-gaps
Field galaxies (total number 150 900)
Muzzin et al. (2013) -band selected COSMOS/UltraVISTA survey 150 900 colour, , mass,
, , SFR
Long GRB host galaxies (total number 52)
Krühler & Schady (2017) , long-duration Swift GRBs detected before 52 colour, , ,
May 2014, part of the GROND 4-hour, TOUGH, SHOALS mass, SFR
BAT-6 samples

Note. – The selection criteria consist of the criteria from each individual survey and those we imposed to build the final samples. All samples were cleaned from duplicates.
We used the re-computed values in Leloudas et al. (2015c).

Table 3: Properties of the comparison samples and their selection criteria
COSMOS/UltraVISTA survey:

To compare SLSN host galaxies to field galaxies, we used the ultra-deep NIR survey UltraVISTA that observed an area of 1.8 deg down to  mag ( confidence). We chose the -band, i.e., mass, selected catalogue by Muzzin et al. (2013) that overlaps with the COSMOS field. This catalogue provides observations in 30 bands from rest-frame UV to NIR. Among all galaxies, we selected those at with SFRs of at least , specific SFRs between and , and “USE” flags equal to one. This sample comprises galaxies with a median redshift of . Because of the small survey area, the number of hosts at is small. This does not affect our analysis because only two SLSNe in our sample are at lower redshifts.

EELGs:

Leloudas et al. (2015c) showed that H-poor SLSNe are preferentially found in EELGs. We built a master sample including results from Atek et al. (2011), Amorín et al. (2014, 2015) and Maseda et al. (2014). Those samples selected EELGs by applying different brightness cuts, colour selection criteria, spectroscopy and redshift constraints. The total sample consists of 227 galaxies with rest-frame [O iii]5007 equivalent widths of  Å between and . All surveys reported stellar mass and SFR for each galaxy, but other properties, such as brightness, colour or , were only reported for certain subsamples.

A summary of the individual surveys and which properties are used in this study is presented in Table 3.

4 Results

4.1 Spectral-energy distribution modelling

4.1.1 Quality of the SED modelling

Figure 2: Selection of spectral energy distributions of hosts of H-poor and -rich SLSNe from 700 to 60000 Å (detections: ; upper limits: ). The solid line displays the best-fit model of the SED. The squares in a lighter shade are the model predicted magnitudes. The fitting parameters are displayed for each SED. See Table 5 and Sect. 3.3 for details. The full collection of SEDs are shown in Figs. 18 and 22.

We made two assumptions to model all SEDs in an automatic and self-consistent way: i) the SEDs can be described by a stellar component with an exponentially declining star-formation history and a contribution from the ionised gas of the H ii regions and ii) the number of filters (n.o.f.) can be reduced to the homogenised filter set in Sect. 3.3. Over 90% of our hosts have good fits with an average of 0.5 and derived physical parameters that are comparable to other galaxy samples (Table 5, Figs. 2, 18, 22).

The fits of only six hosts had between 3.9 and 10.4. The fits of PS1-11bdn and SN1000+0216 are of poorer quality ( and 6.3, respectively) caused by a few data points. The host of PS1-10bzj has very strong emission lines that fall in the wings of the -band transmission function, which increased the normalised to 10.4. Apart from data points in a few individual filters, the fits are nonetheless very good and can be used without restriction.

The fits of CSS100217, PTF11dsf, SN1999bd and SN2006gy have to be used with more caution. Drake et al. (2011a) revealed a narrow-line Seyfert in the host galaxy of CSS100217. Furthermore, Leloudas et al. (2015c) reported on the discovery of broad H and [O iii] in the host spectrum of PTF11dsf, which could be due to an AGN as well. The hosts of SLSNe-IIn SN1999bd and SN2006gy are evolved galaxies that experienced a recent starburst. This is demonstrated by the detection of Balmer lines in both spectra (Smith et al. 2007; Leloudas et al. 2015c; Fox et al. 2015), while the SED cannot be modelled by an exponentially declining star-formation history. A reliable modelling of the SEDs of these three hosts requires a detailed modelling of their star-formation histories and the inclusion of an AGN component, which is beyond the scope of this paper. Leloudas et al. (2015c) mentioned that the host of PTF11dsf could also harbour an AGN. Similar to the three aforementioned hosts, we only use the mass and the -band luminosities of PTF11dsf’s host in our discussion, but not the SFR.

Figure 3: Derived masses (left) and SFRs (right) of galaxies from the spectroscopic sub-sample. The SEDs are fitted with two different procedures: i) the photometry of the galaxies with the contribution of the emission lines is fitted with galaxy templates and an emission line component in Le Phare; ii) the photometry of the same galaxies is fitted after removal of the emission line contribution and switching off the ionised gas component in Le Phare. The values in the upper left corners report the mean bias deviations and the average root square errors (r.m.s.) between the measurements with and without emission-line contribution and their corresponding errors. The solid line indicates the bias between both diagnostics and the dotted lines the mean r.m.s. centred around the bias. The agreement is very good, showing that we can obtain reliable results with Le Phare also for the galaxies where spectroscopic information is not available. The hosts of fast and slow-declining H-poor SLSNe are signified by ‘’ and ‘’, respectively.
Figure 4: Star-formation rates obtained from SED modelling and from emission lines for the spectroscopic sub-sample. The values in the upper left corners report the mean bias deviation and the mean r.m.s. between the H and SED-derived star-formation rates. The solid line indicates the bias between both diagnostics and the dotted lines the mean r.m.s. centred around the bias. Symbols are identical to Fig. 3.

4.1.2 Contribution of emission lines

Our SED modelling includes the contribution of the H ii regions. This is of particular importance because previous studies showed that emission lines can significantly affect the SED fitting (e.g., Castellano et al. 2014; Lunnan et al. 2014; Chen et al. 2015; Santini et al. 2015). This motivated Lunnan et al. (2014) to omit filters that were affected by [O iii]5007, if [O iii] had a large equivalent width, and Chen et al. (2015) to subtract the emission line contribution from the broad-band photometry. Both approaches are strictly limited to objects with host spectroscopy.

Thanks to Le Phare capabilities, we quantify the impact of emission-lines on the SED fitting with a more sophisticated approach. First, we fit the SEDs of the spectroscopic subsample with templates that include a stellar and a gas component. Then, we subtract the contribution of the emission lines from the broad-band photometry and fit the new SEDs with a stellar component only, i.e., the gas component is explicitly switched off in Le Phare.

Figure 3 shows how the primary diagnostics mass and SFR change if emission lines are included in the SED fitting. The absolute value of the average mean bias deviation and the average root mean square error in the mass and SFR estimates are  dex and  dex, respectively, and smaller than the error bars of individual measurements. The most critical object in this analysis is PTF12dam, the most extreme SLSN host galaxy known to date. Its deviations between the mass and SFR estimates with and without lines are , . Apart from this object, the agreement between the two fits is excellent. This reflects the fact that we have good photometry spanning a large wavelength interval and a good handle on the gas emission in the SED fitting, so that the uncertainty in the emission-line contribution does not affect our results.

4.1.3 SED vs. emission-line diagnostics

By combining results from the spectroscopic observations in Leloudas et al. (2015c) with the results from our SED modelling, we have two independent estimates on the recent star-formation activity for our spectroscopic sub-sample. Both diagnostics assume a particular star-formation history and a particular initial mass function. In addition, different diagnostics average the star-formation activity over different time intervals, e.g., the H SFR-indicator is sensitive to the star-formation activity over the past 6 Myr, whereas the SFR derived from rest-frame UV continuum averages over a time period of 100 Myr (e.g., Kennicutt & Evans 2012; Calzetti 2013). Because of the extreme nature of SLSNe, we examine whether we can isolate the differences that occur due to the time-scales that the H and SED-inferred SFRs probe.

Assessing these differences requires that the systematic uncertainties in the data are well understood. Spectroscopic observations with slits are subject to flux losses, because a slit may only cover a part of a given galaxy. Most SLSN host galaxies are relatively compact (Lunnan et al. 2015) so that the expected losses are small. To correct these, Leloudas et al. (2015c) convolved the spectrum of a given object with the filter bandpasses of its imaging data to extract synthetic photometry. In most cases, a simple rescaling was sufficient to adjust the absolute flux scale, i.e., the extracted spectrum is representative for the entire galaxy. Only a few objects required low-order polynomials to correct the warping of the spectrum. In the following, we use the spectroscopic data of a sub-sample of 16 host galaxies with a reliable absolute flux scale.

Figure 4 compares the extinction-corrected SFR’s from SED modelling and H emission lines of these 16 hosts. Both diagnostics reassuringly show consistency. The mean bias deviation and the mean r.m.s. between the H and SED derived SFRs are  dex and  dex, respectively. Conroy (2013) pointed out that a systematic uncertainty in the SED-based SFRs of a factor of 0.3 dex is expected. Our observed value is larger than the expected value but consistent within .

The most interesting object in our sample to identify differences in the SFR indicators is again the host of PTF12dam. Thöne et al. (2015) reported that the head of the tadpole galaxy is characterised by a very young stellar population which is  Myr old. Calzetti (2013) showed that in such cases, the UV SFR estimator will be underestimated by a factor of a few. We measure an excess of  dex in the H inferred SFR. Even in that case, the deviation between the H- and SED-inferred SFRs only has a significance of , reassuring us that even in such an extreme case the SED modelling can provide robust results.

Figure 5: Selection of postage stamps of the hosts of H-poor and -rich SLSN host galaxies in our sample. The images were taken before the SN occurred or after the SN faded. Each panel has a size of where North is up and East is left. The crosshair marks the position of the SNe after aligning on a SN and a host image (H-poor SLSN: blue; H-rich SLSNe: red). If no SN image was available, a circle in blue or red (arbitrary radius) is shown instead, indicating the SN position reported in the literature. The average alignment error is but it exceeds in a few cases. The green circle (arbitrary radius) marks the host galaxy. The observed absolute -band magnitude is displayed in the lower left corner. The image of SNLS07D2bv is smoothed with a Gaussian kernel (width of 1 px) to improve the visibility of the field. The complete collection of postage stamps is shown in Figs. 23 and 25.

4.2 Host offsets

Figures 5, 23 and 25 show postage stamps of each field in our sample. The detected host galaxies (detection rate of ) are marked by green circles. The SN positions, after astrometrically aligning the SN and the host images, are indicated by crosshairs. The average uncertainty of 017 is dominated by the different pixel scales of the SN and host images. In a few examples, this uncertainty exceeds because of the coarse spatial resolution of the SN images, the small spatial overlap of SN and host images, or the low number of reference stars. We lack SN images for 17 hosts in our sample. Their SN positions are indicated by circles as reported in the literature.

Thanks to the high host recovery rate (85% and 100% for H-poor and H-rich SLSNe, respectively), we present a relatively complete distribution of the distances between the SN positions and the barycentres of the host light (predominantly in band) of H-poor and H-rich SLSNe. In addition, we incorporate results on CSS100217 by Drake et al. (2011a), on SN2003ma by Rest et al. (2011) and on Pan-STARRS SLSNe by Lunnan et al. (2015). The observed distribution is skewed to small radii (the expectation value being 1.3 kpc) but has a long tail extending up to 12 kpc. For the smallest offsets, the measurements are comparable to the errors. In this regime, Gaussian noise superimposed on a vector with length results in a non-Gaussian probability distribution of the vector length, i.e., an overestimated host offset (Rice 1944). The expected probability distribution function of a host offset measurement is given by

where is the true offset, is the dispersion of the distribution, which can be assumed to be comparable to the measurement error, and is the modified Bessel function of the first kind. By differentiating with respect to , a closure relation can be derived between the observed offset, its error and the true offset (Wardle & Kronberg 1974):

We solved this equation numerically to build the intrinsic host offset distribution. The black curve in Fig. 6 shows the joint cumulative distribution of H-poor and -rich SLSNe. The grey-shaded regions display the expected parameter space of our distribution after bootstrapping the sample 30 000 times with darker regions, indicating a higher probability. The distribution is well described by the cumulative distribution function of a negative exponential distribution with a mean offset of  kpc.

The fit underpredicts the fraction of hosts with offsets smaller than  kpc and  kpc. The discrepancy for small host offsets can be reconciled with the alignment errors between SN and host image, and intrinsically small host offsets. As the alignment error exceeds the offset measurement, the closure relation is only fulfilled if . Therefore, the fraction of SLSNe with negligible host offsets is a strict upper limit. In addition, any inclination will lead to an underestimation of the true host offset. The blue and red curves in Fig. 6 show the observed offset distribution after separating the sample in H-poor and -rich SLSNe, respectively. Both samples are statistically identical.

The offsets of PTF11rks and SN1999as are  kpc and therefore they exceed the median of 0.7 kpc by a large factor. The host of SN1999as is an irregular galaxy interacting with its environment (Fig. 5). At the explosion site a faint object is detected in continuum. The explosion site of PTF11rks is connected by a linear feature with the nucleus (Perley et al. 2016b). This could point to a spiral galaxy morphology or galaxy interaction whereby the SN exploded in a faint satellite galaxy. Spectroscopic observation of SN1999as by Leloudas et al. (2015c) showed that the explosion site is characterised by strong emission lines. In this case, the true host is a fainter galaxy that is difficult to disentangle from the more massive galaxy.

Figure 6: Host offset cumulative distribution for 41 H-poor (blue) and 13 H-rich (red) SLSNe and the total sample (black). The shaded region displays the expected parameter space after bootstrapping the sample 30 000 times. The dotted, vertical line indicates the median offset. We shifted the distribution by 1 kpc in order to use a logarithmic scaling for presentation purposes.

4.3 Brightness, colour and luminosity

Figure 7: Top: The observed -band host magnitude as a function of redshift for H-poor (blue) and H-rich (red) SLSNe. In case of a -band upper limit, the measurement is displayed as a downward pointing triangle. The hosts of fast and slow-declining H-poor SLSNe are signified by ‘’ and ‘’, respectively, and SLSNe-II by ‘’. Middle: The colour evolution. The observed colour evolution of SUSHIES, GRB host galaxies and star-forming galaxies from the UltraVISTA survey (density plot). Bottom: The colour evolution of galaxies with a metallicity of 0.2 solar for different stellar population ages, derived from templates by Bruzual & Charlot (2003). The tracks are shown up to to avoid corrections for Ly absorption in the host galaxies and in the intergalactic medium. The vectors on the left indicate how extinction, metallicity and emission-lines with very large equivalent widths, such as H and [O iii]5007, can alter the intrinsic colour. Note, H and [O iii]5007 can turn the colour to the blue only at and between and , respectively (indicated by the bars at the bottom).

4.3.1 Brightness and luminosity

Sample Number Mean
redshift (mag) (mag) (mag)
I-fast 11 0.21 (8)
I-slow 5 0.24 (4)
H-poor 27 0.24 (16)
II 3 0.21
IIn 13 0.21 (12) (10) (9)
H-rich 16 0.21 (15) (11) (12)
H-poor 14 0.73 (13) (4)
H-poor 12 1.67 (11) (5)

Note. – The first row of each ensemble property shows the mean value and its error and the second row the standard deviation of the sample. The values of the -band brightness, the -band luminosity and the colour are not corrected for host attenuation. The H-poor and H-rich samples include all SLSNe irrespective of sub-type.
The number of objects with measured colour or with an -band observation are given in parenthesis, if they are less than the total number in the sample.
SNe 1999bd and 2006gy are not considered in the sSFR and SFR calculations because their star-formation histories (SFHs) is more complex than assumed in this paper, while CSS100217 and PTF11dsf are excluded because of a possible AGN contamination.

Table 4: Statistical properties of H-poor and -rich SLSN host galaxies per redshift bin

More than 87% of all hosts were detected at confidence in a -band filter. Their observed distribution, displayed in the upper panel of Fig. 7, extends from  mag (SN2006gy) to  mag (SCP06F6) and shows a clear trend to fainter galaxies as redshift increases (Table 4). The average brightness of SLSN-I host galaxies decreases from  mag at to  mag at , while the dispersion remains at  mag at all redshifts. Compared to a sample of star-forming galaxies from the UltraVISTA survey (density plot in Fig. 7), they are on average fainter and their distributions become more incompatible as redshift increases.

The class of H-poor SLSNe is comprised of fast- and slow-declining SLSNe, which might have different progenitors and host environments. Using the gap in the decline time scale at  days (Table 1), we define a sub-sample of 12 fast and seven slow declining H-poor SLSNe at (Table 1). The properties of the two samples appear to be indistinguishable (Table 4). However, the samples are too small to draw a conclusion yet.

Host galaxies of H-rich SLSNe are on average 1.5 mag brighter than hosts of H-poor SLSNe at (upper panel in Fig. 7; Table 4). Most striking about the SLSN-II/IIn host population is the exceptionally large dispersion of 3.4 mag that is even a factor of 2–3 larger than that of H-poor SLSNe and the UltraVISTA sample (Tables 4, 25; Fig. 25). The large dispersion remains after separating out the three SLSNe-II from the H-rich population (Table 1). The distribution is incompatible with the UltraVISTA sample (chance probability ) and with the fainter and narrower distribution of SLSN-I host galaxies (). Among the hosts of the three SLSNe-II are two of the faintest H-rich SLSN host galaxies in our sample (–26.4; Table 9). They are more than a hundred times fainter than an galaxy at (Faber et al. 2007), and about two magnitudes fainter than the SMC galaxy at .

Panel A of Figure 8 shows the evolution of the absolute -band luminosity (not corrected for host reddening) with redshift. The distribution spans a wide range from to mag. Compared with appropriate luminosity functions (e.g., Faber et al. 2007; Ilbert et al. 2005; Marchesini et al. 2007, tracks in Fig. 8), the span corresponds to a range from a few thousandths of to a few . Clear differences are visible between hosts of H-poor and -rich SLSNe. In their common redshift interval (), the distribution of the H-poor SLSN hosts is narrower by  mag and in addition shifted by  mag towards lower luminosities (Table 4). Intriguingly, the luminosity distribution shows a rapid evolution from at to at . We discuss its origin in Sect. 5.1.

With the -band luminosity distribution in hand we put SLSN host galaxies into context with unbiased GRB and regular core-collapse SN host galaxy samples. Between and , Type I SLSNe reside in galaxies that are  mag less luminous than GRBs. The AD test gives a chance probability of that both distributions are drawn from the same parent distribution (Fig. 14). This result contradicts Japelj et al. (2016), who argued that previously claimed differences between the two populations are an artefact of the comparison methodology. We discuss this finding in Sec. 5.4.1 in detail. The population of SLSN-I host galaxies is also incompatible with those of regular core-collapse SNe from untargeted surveys at all redshifts (; Figs. 14). In contrast, the SLSN-IIn host population is closer to the GRB host population (; Figs. 16).

4.3.2 colour

The middle panel of Fig. 7 shows the redshift evolution of the colour of the 25 H-poor and 11 H-rich SLSN hosts with and -band observations. The colour varies between and 3 mag, though with large errors. No SLSNe are found in extremely red objects (EROs, ). At , SLSN-I hosts are characterised by significantly bluer average colours ( mag; Table 4) than star-forming galaxies from the UltraVISTA survey (grey shaded region;  mag; Table 25). The chance of randomly drawing a distribution from the UltraVISTA sample that is at least as extreme as the SLSN-I is . The average colour is mag bluer and statistically incompatible with those extreme emission galaxies in the VUDS and zCOSMOS surveys (). At , the average colour increases to  mag, but still remains below the average colour of UltraVISTA galaxies (2.43 mag; Tables 4, 25).

The mean colour of hydrogen-rich SLSNe ( mag) is modestly bluer compared to the general population of star-forming galaxies in the UltraVISTA survey and of GRB host galaxies (Tables 4, 25). While the dispersions of the brightness and luminosity distributions are broader than of other galaxy samples, the colour distribution has a dispersion comparable to all other samples [ mag; Tables 4, 25]. Hosts of type II SLSNe tend to be too faint to obtain meaningful -band constraints, which prevents contrasting their properties to the ensemble of type IIn SLSNe.

In the bottom panel of Fig. 7, we overlay expected colour-tracks for the stellar population synthesis templates from Bruzual & Charlot (2003) for a metallicity of 0.2 solar and a wide range of ages. The colour of SLSN-I hosts of  mag at points to stellar population ages of several up to a few hundred million years, whereas H-rich SLSNe are found in galaxies with a redder colour because of more evolved stellar populations. However, the exact relation between colour and age is a complicated function of metallicity, extinction, the equivalent width of emission lines and star-formation histories (for a detailed discussion see Conroy 2013). The vectors in Fig. 7 indicate how they can alter the intrinsic colour.

A critical aspect of this analysis is the and -band observing completeness. Almost all hosts were observed in band, but only were observed in band. The colour incompleteness is a direct consequence of the difficulty to obtain meaningful -band constraints for hosts fainter than –24 mag. This is supported by the SED modelling, which always suggests -band magnitudes below this detection limit and colours that are comparable to the observed colour distribution. In the unlikely case that the hosts without -band observations had –24 mag, the colour distribution would span a range from 0.3 to 4.7 mag. Such red colours are in stark contrast to the observed distribution, the SED modelling, and SN observations (e.g., Quimby et al. 2011c; Inserra et al. 2013; Lunnan et al. 2013; Nicholl et al. 2014).

4.4 Physical properties and distribution functions

Figure 8: Evolution of the physical properties of SLSN host galaxies and comparison samples with redshift. Symbols are identical to previous figures. In panel A, we overlay the evolution of the characteristic luminosity of the -band luminosity function of blue galaxies, reported in Faber et al. (2007), Ilbert et al. (2005) and Marchesini et al. (2007) in grey, and several luminosity tracks. In panel B, we overlay the evolution of the characteristic mass of the mass function from the GAMA (Baldry et al. 2012) and UltraVISTA surveys in grey, and several mass tracks. These characteristic masses and luminosities are defined where the power-law form of the Schechter function cuts off. The parameter space of the UltraVISTA sample is shown as a grey-shaded density plot in panel C. For clarity, measurement errors are omitted for the comparison samples. They are comparable to those of the SLSN host galaxies.
SLSN Redshift
(mag; host) (mag) (mag) (mag) (yr)
SLSN-I host galaxies
CSS140925 0.460
DES14S2qri 1.500
DES14X2byo 0.869
DES14X3taz 0.608
iPTF13ajg 0.740
LSQ12dlf 0.255
LSQ14an 0.163
LSQ14mo 0.256
LSQ14bdq 0.345
LSQ14fxj 0.360
MLS121104 0.303
PS1-10ky 0.956
PS1-10pm 1.206
PS1-10ahf 1.158
PS1-10awh 0.909
PS1-10bzj 0.649
PS1-11ap 0.524
PS1-11tt 1.283
PS1-11afv 1.407
PS1-11aib 0.997
PS1-11bam 1.565
PS1-11bdn 0.738
PS1-12zn 0.674
PS1-12bmy 1.566
PS1-12bqf 0.522
PS1-13gt 0.884
PTF09atu 0.501