The Zwicky Transient Facility Bright Transient Survey. I. Spectroscopic Classification and the Redshift Completeness of Local Galaxy Catalogs

The BTS filter used between 2018 April 1st and 2018 December 31st was deliberately designed to be simple, in order to minimize the risk of false negatives (i.e., real transients that are not saved as part of the program). Using the GROWTH Marshalalert filtering system, we applied the following cuts to the raw ZTF alert stream in order to identify bright SN candidates:

• 1.  
  Alerts at low Galactic latitudes ($\left|b\right|\leqslant 7^\circ$) are rejected.
• 2.  
  Alerts with a random-forest-based machine-learning real-bogus score (rbscore; Mahabal et al. 2019) of less than 0.2 are rejected. This choice results in a completeness (i.e., 1—false negative rate) of >99% (Figure 16 in Duev et al. 2019).
• 3.  
  Alerts produced at the position of known stars, as identified in the catalog created by Tachibana & Miller (2018), are rejected. A small fraction of galaxies (estimated to be <0.5% in Tachibana & Miller 2018), and thus nuclear SNe, will be missed as a result of this cut.
• 4.  
  Alerts produced close to very bright stars have been rejected (<20'' for <15 mag stars; ∼1% loss of survey area²⁹ ).
• 5.  
  Alerts that do not include at least two detections separated by a minimum of 30 minutes are rejected (moving object filter).
• 6.  
  Alerts with negative flux relative to the reference image are rejected.
• 7.  
  The alert must include at least one epoch with m < 19 mag (in the g_ZTF or r_ZTF band), otherwise it is rejected.

The BTS filter effectively passes all alerts brighter than 19 mag that are not consistent with stellar events (star detected in PS1) or moving objects for human vetting (scanning; Section 2.2).

On a typical night in 2018, a few hundred alerts passed the BTS filter. These BTS transient candidates were visually inspected by a team of scanners on a daily basis. To organize this effort, we use the GROWTH Marshal (Kasliwal et al. 2019), where the lightcurves and image cutouts (science, reference, and subtraction) contained in the Avro packets of the passing alerts are collected on a scanning page for each night. Supplementary information is also displayed, such as PS1 and SDSS color composite cutouts centered on the position of the transient; the star–galaxy separation score (sgscore), which gives a probability that the closest PS1 counterpart is an extended galaxy or a point-like star (Tachibana & Miller 2018); multiband photometry of this PS1 counterpart, a cross-check for known near earth objects (NEOs); and information about if and when there have been previous ZTF alerts at the same position that are not part of the 30 day history contained in the alert packet itself. External catalog cross-matches (e.g., NED, TNS, SIMBAD, VizieR) are also linked through the GROWTH Marshal to provide additional contextual information for each potential SN candidate.

The human-vetting process essentially consisted of inspecting the information contained in the Avro packets for each alert that passed our filter and also taking into account any relevant supplementary information available in order to rule out variability from a stellar counterpart, and to reject alerts produced by known AGNs.

Among the passing alerts, 5–15 SN candidates are typically identified by the human scanners as good SN candidates and assigned for spectroscopic follow-up, per night.³⁰ The two main contaminants in our scanning process are cataclysmic variables (CVs) that are too faint to be seen in PS1 in their quiescent phase, and therefore lack an sgscore, and AGNs. Both of these must be avoided, given our limited spectroscopic resources. We have found that the vast majority of CVs can be avoided by monitoring the lightcurve behavior until ∼1 week after the initial outburst and comparing the evolution with typical CV lightcurves.³¹ This is especially effective as ZTF produces both g_ZTF- and r_ZTF-band photometry for virtually all transients that are detected.

Filtering AGNs is more challenging: excluding all AGNs from the BTS could inadvertently reject an SN that has exploded near the nucleus of a galaxy that harbors an AGN. Filtering on past variability is a very effective way of excluding AGNs, but for the sample presented here, the baseline of ZTF observations was only a few weeks or months, which is not always a sufficient amount of time for the AGN to change in flux enough to be recognized as a variable object. We have generally not triggered spectroscopic follow-up for alerts that are positionally coincident with known AGNs (e.g., the ALLWISE mid-IR AGN catalog, Secrest et al. 2015; the Million Quasar catalog, Flesch 2015), unless the photometric evolution of the transient is very similar to that of an SN. Thus, the BTS is incomplete for SNe near AGNs (a focused survey with the specific goal of discovering SNe in galaxies with an AGN is needed to address this).

2.2.1. Completeness of the BTS Filter

The primary classification instrument of BTS candidates is the SEDM (Blagorodnova et al. 2018; Rigault et al. 2019). For candidates that pass the BTS filter (Section 2.1) and visual inspection (Section 2.2), we assign spectroscopic SEDM observations with the following priorities: transients that are, or are likely to become, brighter than 18.5 mag are scheduled with the highest priority (P3). Transients that appear likely to peak between 18.5 and 18.75 mag are assigned a lower priority (P2), and transients expected to peak fainter than 18.75 mag are triggered at the lowest priority (P1). For a typical BTS source (m ≈ 18.5 mag) observed under typical observing conditions for 1800 s, SEDM obtains a signal-to-noise ratio (S/N) ≈ 12 per resolution element and an integrated S/N ≈ 50 in the region covered by the r_ZTF filter.

The SEDM queue is designed so that lower priority targets (e.g., P1, P2) are only observed if no higher priority targets are present in the queue that could be observed within the same observation-time window. The purpose of this priority scheme is twofold. First, it ensures that we can reach a high level of completeness for m_peak < 18.5 mag sources by allowing some margin for error on the fainter end of 18.5 mag. Second, when the queue allows (e.g., periods of consistently good weather), significantly fainter targets, including those as faint as ∼19 mag, can also be observed in addition to our high-priority (P3) targets.

By default, triggers that enter the SEDM queue remain active for 7 days. If the transient has not been observed at this time, the candidate is reassigned to SEDM or to larger telescopes if the candidate has become too faint for SEDM (m ≳ 19 mag). Larger telescopes are also used if classification with SEDM is unsuccessful, which typically happens only if the candidate was observed in poor sky conditions, there is strong host-galaxy contamination, or if higher resolution is needed for a secure classification. For this purpose, we have primarily used P200 and Keck I, but supporting observing programs at LT, NOT, and APO have contributed significantly as well. Community efforts (e.g., ePESSTO; Smartt et al. 2015) have also contributed through TNS (see Section 3.1 for details).

2.3.1. Spectroscopic Completeness

As previously noted, spectroscopic observations of BTS SN candidates are primarily obtained with SEDM. In cases where the candidates were too faint to be observed with SEDM, or scheduling conflicts prevented SEDM observations, or the SEDM spectra proved to be ambiguous, spectra were obtained with spectrographs on larger aperture telescopes: the Double Spectrograph (DBSP; Oke & Gunn 1982) on P200, the Low Resolution Imaging Spectrometer (LRIS; Oke et al. 1995) on the Keck I telescope, the Dual Imaging Spectrograph (DIS) on the APO 3.5 m telescope, the Spectrograph for the Rapid Acquisition of Transients (SPRAT; Piascik et al. 2014) on the 2.0 m LT, and the Alhambra Faint Object Spectrograph and Camera (ALFOSC) on the 2.56 m NOT.

SEDM data were reduced by the pipeline described in Blagorodnova et al. (2018) for data until 2018 August and by the automatic PYSEDM pipeline described in Rigault et al. (2019) for data after 2018 August. DIS data were reduced with the pyDIS package (Davenport et al. 2016). DBSP data were reduced using the PyRAF-based (Science Software Branch at STScI 2012) pipeline pyraf-dbsp (Bellm & Sesar 2016). ALFOSC data were reduced using standard procedures and tools based on IRAF (Tody 1986). LRIS data were reduced using the LRIS automated reduction pipeline (LPipe; Perley 2019).

For SN candidates where we could not obtain spectroscopic observations from any of the above-mentioned telescopes, public TNS spectra were analyzed from the following instruments: the Asiago Faint Object Spectrograph and Camera (AFOSC) on the Asiago Ekar 182 cm telescope, FLOYDS on the Faulkes Telescope South (FTS) in Australia operated by Las Cumbres Observatory (LCO), the Wide Field Reimaging CCD (WFCCD) on LCO's duPont telescope, the ESO Faint Object Spectrograph and Camera v.2 (EFOSC2) on the New Technology Telescope (NTT), the Dolores (Device Optimized for the LOw RESolution) on the Telescopio Nazionale Galileo (TNG), the Intermediate-dispersion Spectrograph and Imaging System (ISIS) on the William Herschel Telescope (WHT), and the DeVeny spectrograph on the Discovery Channel Telescope (DCT).

Preliminary classifications are made via SuperNova IDentification (SNID; Blondin & Tonry 2007) template matching and visual inspection. SNID is automatically applied to all SEDM spectra, whereas for all other instruments, SNID is applied to the spectra by the observer. These preliminary classifications are annotated and recorded within the BTS program on the GROWTH Marshal and subsequently sent to TNS within 1 to 2 days.³⁶

We revisit each of the preliminary classifications for this study in order to develop a homogeneous classification scheme. For this purpose, we developed a custom process to spectroscopically classify the 761 SNe in the BTS sample. For each BTS SN, we identified the top 15 spectral matches (rlap_min ≥ 5) from SNID to the most recent spectrum available on the GROWTH Marshal. We used the latest spectrum from the Marshal under the assumption that BTS targets only received additional spectroscopic observations when the initial classification was inconclusive. The SNID templates used for this process include the developer defaults,³⁷ as well as SNe Ia and a few non-SN templates from the Berkeley SN Ia program (BSNIP; Silverman et al. 2012); SN Ib/c templates from Modjaz et al. (2014), Liu et al. (2016), Modjaz et al. (2016), and Williamson et al. (2019); and SN IIP templates from Gutiérrez et al. (2017).

Following the SNID matching, we produced plots showing a comparison between the observed BTS spectrum and the (redshift-corrected) template spectrum from SNID. These plots were visually inspected to identify the best-matching template. In practice, the sample was split into six groups, and each group was inspected by a member of our team (C.F., A.A.M., A.D., Y.S., K.T., A.G.). While identifying the best-matching template, we recorded the SN type and redshift, as well as the name and phase of the template SN spectrum. In cases where the same classification was reported for all 15 matches, we recorded the type, redshift, and phase from the top match from the SNID output. We otherwise selected the best match based on common prominent SN spectral features (H, He, Si, Ca, Fe, etc.). If the top 15 matches from SNID proved ambiguous, we used either the ZTF lightcurve or alternative spectra to remove the ambiguity. For example, in cases with multiple matches to both SNe Ia and SNe Ic, the telltale secondary near-infrared (nIR) peak of SNe Ia can typically be seen in ZTF r_ZTF-band lightcurves. The secondary nIR peak is unique to SNe Ia, as explained in Kasen (2006). It occurs following a recombination transition of the iron-group elements in the ejecta, whereby the strength of the Fe iii and Co iii lines decreases, and there is a corresponding strengthening of the Fe ii and Co ii lines (see also Blondin et al. 2015). While these papers have mostly considered λ ≳ 7500 Å, a related "shoulder" in the r-band lightcurve is also observed at approximately the same time (Papadogiannakis et al. 2019a, 2019b).

If at this stage a classification still proved ambiguous, then the SN was examined by another member of the team. For consistency, a final check of all ambiguous classifications was performed by two members of the team (C.F., A.A.M.). Ultimately, we have classified 761 SNe via their spectra and lightcurves (Table 1). Out of these, 503 were classified using SEDM spectra, 86 using P200-DBSP, 76 using Keck I-LRIS, 20 using LT-SPRAT, 11 using APO-DIS, and 9 using NOT-ALFOSC. Finally, 56 were classified based on publicly available spectra on TNS.

We note that the positions reported in Table 1 are obtained by taking the weighted average of the position of the SN in every image in which the SN is detected (i.e., for every alert associated with the SN). The updated positions are more accurate than those reported to TNS, which typically only include a single low-S/N detection of the SN.

We broadly classify all BTS SNe as belonging to one of four different classes: SNe Ia, SNe II, SNe Ib/c, and superluminous SNe (SLSNe). As detailed above, these classifications are primarily made via the SN spectra; however, in some cases the photometric evolution of the SN also informs the classification. This is especially true of the SLSNe, which are defined by their luminosity (typically M < −21 mag, Gal-Yam 2012; although ZTF adopts M < −20 mag).

Of the 761 SNe, we find that 547 are SNe Ia, 155 are SNe II, 40 are SNe Ib/c, and 19 are SLSNe. The fraction of discoveries belonging to each of these classes is in agreement with the results from Li et al. (2011) for a magnitude-limited survey, as shown in Figure 3.³⁸ Figure 3 also shows the relative rate of SNe found by ASAS-SN (Holoien et al. 2017a, 2017b, 2017c, 2019), which, like the ZTF BTS and unlike LOSS, does not target specific galaxies when searching for transients.³⁹ Both ZTF and ASAS-SN find a higher fraction of SNe II than LOSS, although these estimates all agree to within the uncertainties. By targeting massive galaxies, including a significant fraction of passive elliptical galaxies, LOSS may have been slightly biased against finding CC SNe (see Taubenberger 2017 and references therein). The relative rate of SLSNe is somewhat higher for ZTF compared to ASAS-SN, but still consistent within the uncertainties (see Section 3.3.4).

Zoom In Zoom Out Reset image size

Table 2 summarizes the relative fraction of SNe in each class for the ZTF BTS, ASAS-SN, and LOSS. For LOSS, we directly use the estimates from Li et al. (2011), while for the BTS and ASAS-SN, we assume the observations are drawn from a multinomial distribution and estimate 95% confidence intervals on the true rate via the approximate method of Goodman (1965) as implemented in the MultinomCI (Signorell et al. 2019) package in R. The true uncertainties on these fractions require a detailed estimate of the completeness of the BTS, which is beyond the scope of this paper, and will be addressed in future work (J. Nordin et al. 2020, in preparation).

Table 2.  Relative SN Rates

Survey	${ \mathcal R }$(Ia)	NIa	${ \mathcal R }$(II)	NII	${ \mathcal R }$(Ibc)	NIbc	${ \mathcal R }$(SLSN)	NSLSN
LOSS	$0.792{\pm }_{0.055}^{0.044}$	⋯	$0.166{\pm }_{0.039}^{0.050}$	⋯	$0.041{\pm }_{0.013}^{0.016}$	⋯	⋯	⋯
ASAS-SN	$0.742{\pm }_{0.045}^{0.040}$	607	$0.211{\pm }_{0.037}^{0.043}$	173	$0.043{\pm }_{0.016}^{0.024}$	35	$0.004{\pm }_{0.003}^{0.012}$	3
ZTF BTS	$0.719{\pm }_{0.048}^{0.043}$	547	$0.204{\pm }_{0.038}^{0.043}$	155	$0.053{\pm }_{0.018}^{0.027}$	40	$0.025{\pm }_{0.012}^{0.021}$	19

Note. Relative rates ${ \mathcal R }$ of SNe Ia, II, Ibc, and SLSN from a flux-limited search. Results for LOSS are taken directly from Table 7 in Li et al. (2011) for a 1 day cadence and are based on an assumed luminosity function and Monte Carlo simulations. Thus, the total number of each type of SN is not relevant and therefore not reported. Results for ASAS-SN use all discovered and recovered SNe reported in Holoien et al. (2017a, 2017b, 2017c, 2019). Uncertainties for both ASAS-SN and the ZTF BTS include 95% confidence intervals (see text).

Download table as:  ASCIITypeset image

As a test of our classification accuracy, we compare our final classifications to those made by ePESSTO (Smartt et al. 2015).⁴⁰ There are 26 sources in common between our BTS classifications and those made by ePESSTO, and the classifications are in agreement for all but two sources: ZTF18abmrhom (SN 2018ffi) and ZTF18abtswjk (SN 2018gfx). We classify ZTF18abmrhom as an SN Ia, whereas the ePESSTO spectrum is classified as having just galaxy light. The ZTF spectrum of ZTF18abmrhom, which was obtained two nights prior to the ePESSTO spectrum, shows clear and strong Si ii absorption. The same broad feature can be seen in the ePESSTO spectrum, which is otherwise dominated by emission from the host galaxy. Furthermore, the r_ZTF-band lightcurve shows a "shoulder" a few weeks after maximum light. Taken together, it is very likely that ZTF18abmrhom is an SN Ia. The classification of ZTF18abtswjk is more challenging: the spectrum clearly shows narrow emission lines, and we have classified the event as an SN IIn, whereas ePESSTO classified the spectrum as an AGN. The photometric evolution of ZTF18abtswjk is far more reminiscent of SNe than AGNs: the transient exhibits a ∼30 day rise, followed by a very slow, monotonic decline (the transient has exhibited a monotonic fade by ∼1.5 mag over the period covering 2018 and 2019) as is characteristic of many SNe IIn (e.g., Turatto et al. 1993). Furthermore, the WISE colors for the host galaxy are not consistent with AGNs (Jarrett et al. 2011). Thus, based on our comparison to ePESSTO, we conclude that our classifications are of high fidelity.

There were three BTS sources for which we attempted spectroscopic classification, but the nature of the transients remains unresolved. Each of the three candidates are positionally coincident with the nucleus of their host galaxies. ZTF18aaqkdwf (SN 2018fhd) exhibits a broad rise and decline over ∼200 days and significant P-Cygni-like feature around Hα. However, variability at the location of ZTF18aaqkdwf was detected ∼3.5 yr prior to the 2018 variability by iPTF, and the host galaxy is a bright point source in the radio (Helfand et al. 2015). Thus, the variability may be due to an AGN. The lightcurve of ZTF18abuqhje (SN 2018gki) has poor coverage with only four r_ZTF detections that show a decline of ∼0.7 mag over ∼15 days. We obtained two spectra of ZTF18abuqhje, both of which show many narrow lines that could be consistent either with an AGN or an SN IIn. However, we detected variability from this nucleus ∼2.5 yr prior to 2018 with iPTF, which would be consistent with an AGN. The general evolution of the ZTF18aarcchg (AT 2018boa) r_ZTF lightcurve is consistent with an SN: the transient rises by ∼1.5 mag over ∼20 days, followed by a monotonic decline. Spectra of ZTF18aarcchg exhibit a blue continuum with a superposed narrow emission from the Balmer series, [O iii], and [S ii], all of which are consistent with star formation (the host is classified as a star-forming galaxy; Maraston et al. 2013). Without discernible SNe features in the spectrum, we cannot classify this transient. We exclude these three candidates from the subsequent analysis.

The primary purpose of this study is to improve the measurement of the RCF and to extend the redshift coverage of this measurement beyond what is presented in Kulkarni et al. (2018). To that end, we only need to separate SNe Ia from all other transients in the ZTF BTS. Nevertheless, we attempt to make subtype classifications, based on the SNID matches described above, as detailed for each class below. We caution, however, that the vast majority of these classifications are made with very low-resolution (R ≈ 100) SEDM spectra, and as a result, there are significant uncertainties on the subclass of any individual SN. Furthermore, any biases in the SNID template set can influence the final BTS classifications. The SNID templates were compiled from heterogeneous sources and do not perfectly reflect the discovery space of an untargeted transient survey, meaning the SNID templates themselves are a further source of uncertainty for the final subclassifications presented in Table 1 (see Blondin & Tonry 2007 for further details).

3.3.1. SNe Ia

3.3.2. SNe II

3.3.3. SNe Ibc

3.3.4. SLSNe

In addition to providing SN spectral types, SNID estimates the redshift of the SN it is attempting to classify. We adopt the redshift of the best-matching SNID template as the redshift of the SN, z_SN.⁴² The redshift distribution for BTS SNe is shown in Figure 4, where we adopt the redshift of the host galaxy (z_host) when known; otherwise, we show z_SN.

Zoom In Zoom Out Reset image size

We can estimate the accuracy of the SNID redshift measurements using the subset of BTS SNe that have host galaxies with known redshift (for more on the identification of BTS host galaxies, see Section 4.2). We find that z_SN is a good estimator of the host-galaxy redshift, z_host, as summarized in Figure 5 for the 345 ZTF BTS host galaxies with known redshifts.

Zoom In Zoom Out Reset image size

The main panel in Figure 5 shows the difference between z_SN and z_host as a function of z_host. The residuals show that for SNe Ia, there is a relatively small scatter (σ ≈ 0.0037)⁴³ and virtually no bias in the estimates of z_SN. The scatter is somewhat higher for CC SNe (σ ≈ 0.0047), where SNID appears to systematically overestimate the true redshift, as is seen in the Gaussian kernel density estimate (KDE) shown in the right panel of Figure 5. We find some evidence for an increased scatter at higher redshifts. For SN Ia hosts with z ≤ 0.04, σ ≈ 0.0032, while for hosts with 0.08 ≤ z < 0.12 σ ≈ 0.0045 (though the 90th percentile widths are nearly identical in these two regions). This increased scatter makes sense as the S/N typically decreases for higher redshift SNe; for SNe outside the redshift range shown in Figure 5, the typical uncertainty on any individual redshift may be larger than ∼0.004. We also find that the scatter is not appreciably larger when restricting the sample to only those SNe that have been observed by SEDM. In the analysis that follows, we assume that z_host = z_SN for normal SNe Ia, and that the typical uncertainty on z_SN is 0.004. We note that (i) this is very consistent with the uncertainty (σ = 0.005) reported in Figure 19 in Blondin & Tonry (2007), even though a very different redshift range was used (z = 0.1–0.8), and (ii) our estimated redshift uncertainty on z_SN is much larger than the uncertainty in the wavelength calibration of SEDM (Figure 13 in Rigault et al. 2019). These two findings indicate that the accuracy of our SNID-based redshifts (z_SN) is not limited by the low resolution of SEDM.

Correctly associating a newly discovered transient with its host galaxy is a challenging problem, especially when the redshift of the host candidate is unknown. Simply identifying the closest galaxy (in angular offset) is likely to produce a significant number of misidentifications, especially in the case of nearby SNe for which angular offsets relative to the host nuclei may be quite large.

We use a combination of automated procedures and visual inspection to identify hosts for ZTF BTS SNe. As an initial pass, we query the NASA Extragalactic Database (NED)⁴⁴ for all galaxies within 2' of the SN position. Within this list, the galaxy with the smallest angular separation from the SN and a cataloged redshift in NED is automatically assigned as the host. In cases where there are no cataloged galaxy redshifts within 2' of the SN, the galaxy with the smallest angular separation from the SN is assigned as the host. In cases where the z_host and z_SN significantly differ ($\left|{z}_{\mathrm{SN}}-{z}_{\mathrm{host}}\right|\gt 0.05$), the galaxy with the smallest angular separation from the SN is assigned as the host. From here we calculated the projected separation, d_p, between the SN and the host galaxy using the redshift of the SN (see above). We visually inspect host candidates with a projected separation d_p ≥ 19 kpc. In most of these cases, it is clear that the automated procedure identified a background galaxy that is clearly not the host, in which case we update the host with the NED galaxy with the smallest angular separation from the SN. Following this procedure, there were a total of 12 SN host identifications with separations d_p ≥ 19 kpc (9 SNe Ia and 3 SNe II). In 11 of these 12, the host redshift is known and that redshift matches that of the SN, providing confidence in these associations. We cannot rule out the possibility that these SNe occurred in faint dwarf galaxies that are associated with the bright galaxy that has been identified as the host. For the purposes of the RCF calculation below, we assume each of these identifications to be correct. For the remaining SN, ZTF18acrcetn (SN 2018jag), there is a very bright (r' = 14.3 mag) elliptical galaxy, PSO J015.9596+10.5902, in the field of the SN. The SN is ∼3.2 Petrosian radii from the galaxy, which has an SDSS photometric redshift, 0.052 ± 0.012 (Abolfathi et al. 2018), that agrees with the SN redshift determined by SNID, 0.053 (which gives d_p ≈ 24 kpc). For the calculations below, we assume that this identification is correct.

Following this procedure, we use deep i-band stack images from PS1 to visualize the position of each SN relative to its host galaxy. In the vast majority of cases, these images confirmed a clear association between the SN and host galaxy. In some cases, the putative host was extremely faint and a significantly brighter galaxy with only slightly larger angular separation was selected as the likely host. Finally, there were a handful of cases where the association was ambiguous or with a low-S/N PS1 detection. We find that for 40 SNe in the sample, the host identification is ambiguous. We exclude these SNe from the host-galaxy analysis below. We thus identify host galaxies for 721 of the 761 SNe in the ZTF BTS sample. Properties of the BTS SNe host galaxies are summarized in Table 3, including notes on each of the SNe identified as having ambiguous hosts. For the 40 ambiguous cases, we find that 29 SNe have no discernible host, including 17 SLSNe, for which host galaxies are typically not found in imaging at the depth of PS1 in this redshift range (e.g., Quimby et al. 2011). The remaining 11 are either roughly equidistant between multiple galaxies of the same brightness, or very close to a faint galaxy, with a significantly brighter galaxy at a similar redshift of the SN residing at much larger angular separation.

Table 3.  BTS Host Galaxies

ZTF	Host	αhost	δhost	zhost	SN Offset	dp	mg	mr	mi	mz	my	mW1	mW2	E($B-V$)a
Name	Name	(J2000.0)	(J2000.0)		(arcsec)	(kpc)	(mag)	(mag)	(mag)	(mag)	(mag)	(mag)	(mag)	(mag)
ZTF18aabssth	PSO J165.1878+22.2877	11:00:45.07	+22:17:15.8	0.02291	4.44	2.06	16.89	16.49	16.38	16.38	16.61	15.79	16.38	0.015
ZTF18aabxlsv	PSO J157.4639+09.0106	10:29:51.32	+09:00:38.1	0.04797	9.53	8.96	16.17	15.48	15.27	15.02	15.02	14.82	15.30	0.025
ZTF18aaemivw	PSO J158.4280+39.4908	10:33:42.72	+39:29:26.8	0.06807	0.33	0.43	16.66	16.03	15.67	15.53	15.35	14.94	15.08	0.012
ZTF18aagpzjk	PSO J119.8484+16.4214	07:59:23.61	+16:25:16.9	0.01631	26.83	8.91	14.96	14.49	14.33	14.49	14.87	13.85	14.43	0.031
ZTF18aagrdcs	PSO J218.3331+41.2658	14:33:19.95	+41:15:56.9	0.01814	5.43	2.00	18.53	17.62	17.45	18.11	18.86	17.32	18.00	0.012
ZTF18aagrtxs	PSO J198.6087+50.9792	13:14:26.08	+50:58:45.0	0.02966	7.94	4.72	⋯	⋯	⋯	13.77	13.68	13.70	14.36	0.010
ZTF18aagstdc	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯	⋯
ZTF18aagtcxj	PSO J248.0475+42.7139	16:32:11.40	+42:42:50.0	0.03240	2.40	1.55	16.72	15.70	15.31	15.00	14.82	14.28	14.75	0.011
ZTF18aahesrp	PSO J128.9390+28.2705	08:35:45.35	+28:16:13.7	⋯	1.36	1.35	19.43	18.99	18.82	18.69	18.68	19.14	19.86	0.036
ZTF18aahfeiy	PSO J154.3116+43.5219	10:17:14.78	+43:31:18.8	0.07126	10.09	13.71	17.13	16.69	16.38	16.43	16.50	16.19	16.61	0.010
ZTF18aahfgyz	PSO J175.2822+24.8218	11:41:07.72	+24:49:18.5	0.01201	8.88	2.18	⋯	⋯	⋯	⋯	11.84	11.98	12.67	0.023
ZTF18aahfqbc	PSO J182.6357+39.4059	12:10:32.58	+39:24:21.1	0.00332	73.35	5.03	12.09	11.52	⋯	⋯	⋯	9.98	9.51	0.024
ZTF18aahfxnn	PSO J176.5486+09.3548	11:46:11.67	+09:21:17.1	⋯	4.35	6.12	18.89	18.49	18.20	18.22	18.09	18.09	18.69	0.026
ZTF18aahfzea	PSO J181.3610+20.3031	12:05:26.64	+20:18:11.1	0.02373	14.29	6.84	⋯	19.03	19.86	⋯	⋯	⋯	⋯	0.033
ZTF18aahheaj	PSO J195.1118+18.6197	13:00:26.83	+18:37:11.1	0.05272	4.74	4.87	18.24	17.79	17.73	17.95	17.82	16.69	17.07	0.022
ZTF18aahhenr	PSO J207.3615+26.4619	13:49:26.75	+26:27:42.7	0.07636	5.64	8.16	17.04	16.21	15.80	15.68	15.66	15.29	15.80	0.010
ZTF18aahhqih	PSO J138.7320+46.9031	09:14:55.68	+46:54:11.2	0.01425	5.22	1.52	14.06	13.53	⋯	14.07	13.44	12.62	13.09	0.016
ZTF18aahhzqn	PSO J178.7588+32.0746	11:55:02.11	+32:04:28.7	0.03106	12.34	7.67	15.55	15.07	14.87	14.71	14.64	14.77	15.23	0.020
ZTF18aahjafd	PSO J198.2744+23.4593	13:13:05.86	+23:27:33.5	0.03588	0.21	0.15	18.15	17.54	17.17	16.99	16.89	17.34	18.02	0.010
ZTF18aahmhxu	PSO J176.7668+19.5508	11:47:04.03	+19:33:02.9	⋯	0.62	1.77	19.77	18.98	18.54	18.40	18.24	18.06	18.38	0.027

Notes. Optical grizy photometry is taken from PS1 Kron magnitude measurements, while mid-IR W1 and W2 photometry are from Tractor and WISE (see text). All magnitudes are reported in the AB system, and no correction for extinction has been applied. The host galaxies for ZTF18acbzoyh (SN 2018hqu) and ZTF18acdwohd (SN 2018ids) are not detected in the PS1 catalog (see text).

^aE($B-V$) is determined using the Schlafly & Finkbeiner (2011) updates to the Schlegel et al. (1998) maps.

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

The host-galaxy coordinates available via NED come from a heterogeneous set of catalogs and surveys, resulting in an astrometric offset between NED galaxy coordinates and ZTF SN positions, which are measured relative to Gaia (Gaia Collaboration et al. 2016). We crossmatch the host positions against the PS1 catalog, which is also calibrated against Gaia, in order to place the BTS SNe and host galaxies on the same relative astrometric system. A total of 715 of our initial host positions have counterparts within 2'' in the PS1 DR1 MeanObject table, which is astrometrically calibrated against Gaia. Four of the hosts, those associated with ZTF18aapgrxo (SN 2018bym), ZTF18aayjyub (SN 2018cod), ZTF18acaeous (SN 2018hbu), and ZTF18acrknyn (SN 2018jef), are too faint to be included in the PS1 MeanObject table, and instead, we use PS1 positions from the StackObjectThin table. The last two hosts, associated with ZTF18acbzoyh (SN 2018hqu) and ZTF18acdwohd (SN 2018ids), are not detected in the PS1 catalog, and we instead use their positions from SDSS and the Galaxy Evolution Explorer (GALEX; Martin et al. 2005), respectively, in Table 3. Following this update of the positions, we recalculate the distribution of host-galaxy separations, which is shown in Figure 6. We find a nearly identical distribution in projected separation for CC SNe and SNe Ia (middle panel of Figure 6). We use a two-sample Kolmogorov–Smirnov (KS) test and a χ² test for independence to determine the statistical difference between the two distributions and find no significant difference between the projected offsets of CC SNe and SNe Ia.

Zoom In Zoom Out Reset image size

Using the newly identified host offsets, we can additionally examine whether or not there is a strong bias against finding nuclear SNe in ZTF, as has been found for other surveys (see Holoien et al. 2019). Figure 7 compares the cumulative distribution of angular offsets for bright (m_peak ≤ 17 mag) BTS SNe and ASAS-SN.⁴⁵ We find remarkably similar distributions between ZTF and ASAS-SN. Both a two-sample KS test and a χ² test for independence do not show a statistically significant difference between the two samples. This suggests that ZTF is no more biased against finding nuclear SNe than ASAS-SN, in contrast to other surveys (e.g., Holoien et al. 2019). We caution that while the relative bias is comparable to ASAS-SN, both ASAS-SN and ZTF may have an unmeasured absolute bias against finding SNe in the nuclei of galaxies.

Zoom In Zoom Out Reset image size

Following host identification, we need to determine the absolute magnitude of the host galaxies in order to measure the RCF (see Kulkarni et al. 2018 and below for further details). In this study, we focus on the mid-IR flux of the host galaxies, primarily for two reasons: (i) a galaxy's mid-IR absolute magnitude serves as a good proxy for the total galactic stellar mass (Wen et al. 2013), and (ii) mid-IR photons are mostly transparent to dust in the Milky Way, meaning that significant reddening corrections are not needed to estimate a galaxy's absolute magnitude in the mid-IR. The WISE (Wright et al. 2010) satellite conducted an all-sky survey in the mid-IR, and we use WISE images to determine the brightness of ZTF BTS SN host galaxies at 3.4 μm.

The largest WISE source catalogs (e.g., unWISE; Schlafly et al. 2019) utilize an unresolved (i.e., stellar) point spread function (PSF) to measure source flux. This PSF is not appropriate for many of the low-z galaxies in our sample, which are clearly resolved in WISE imaging. The recent development of The Tractor (Lang et al. 2016a) enables "forced" WISE flux measurements, where the mid-IR apertures are determined via sources detected in (relatively) high-resolution optical images. The Tractor was used to measure the mid-IR flux of ∼400 million WISE sources that were detected by SDSS (Lang et al. 2016b) and now is also being applied to Legacy Survey images (Dey et al. 2019). There are 371 BTS host galaxies that have Tractor forced mid-IR photometry from both Legacy Survey and SDSS images, while an additional 157 hosts have detections in just Legacy Survey images and another 90 hosts have detections in just SDSS⁴⁶ (hereafter we refer to this aperture-matched forced photometry as Tractor photometry). We include Tractor photometry based on both Legacy Survey and SDSS images in our analysis of the RCF below. For the 371 sources detected both in the Legacy Survey and SDSS, we compare the Tractor photometry derived from each set of images and measure a sample standard deviation of 0.008 in Δ flux/flux. This small difference suggests that there are no systematic effects introduced by combining photometry from the two different optical catalogs. Ultimately, this results in 618 host galaxies with 3.4 μm flux measurements that we can use in the analysis of the RCF. Including only the brightest mid-IR galaxies in our sample will bias our final measurement of the RCF, as discussed below.

To date, SDSS has been the most prolific survey in terms of spectroscopically measuring galaxy redshifts, with ∼2.8 million cataloged galaxies and counting (Aguado et al. 2019). At this stage, significant improvements to the RCF will require tens of millions of new redshift measurements. Fortunately, within the next few years, we will enter the era of supremely multiplexed spectrographs (e.g., the 4 m Multi-Object Spectroscopic Telescope (4MOST),⁵⁰ the Dark Energy Spectroscopic Instrument (DESI),⁵¹ the Subaru Prime Focus Spectrograph (PFS),⁵² Euclid,⁵³ the Wide Field Infrared Survey Telescope (WFIRST)⁵⁴ ), which could dramatically increase the number of galaxies with known spectroscopic redshifts within the local universe. The best strategy to this end is to obtain spectra of bright galaxies with currently unknown redshifts, as is planned as part of the DESI Bright Galaxy Survey (DESI Collaboration et al. 2016). Using the same methodology described above, we can estimate the RCF as a function of galaxy brightness (as traced by the PS1 r band), rather than z or M_W1. The results from this exercise are summarized in Figure 9.

Zoom In Zoom Out Reset image size

From Figure 9, it is clear that the vast majority of extremely bright galaxies (r_PS1 ≲ 14 mag) have cataloged redshifts. Figure 9 also shows the RCF for galaxies that are and are not within the SDSS imaging footprint,⁵⁵ and a comparison of these two curves highlights the crucial role that SDSS played in terms of identifying galaxies in the local universe and spectroscopically measuring their redshifts. SDSS pushes the completeness of redshift catalogs ∼2 mag fainter than what is observed outside the SDSS footprint. From our analysis, it is possible to estimate the completeness of existing redshift catalogs as a function of depth, and we find that catalogs are currently ∼90%, 50%, and 10% complete to a depth of r_PS1 ≈ 14.7 mag, 16.9 mag, and 19.1 mag, respectively. Because the DESI bright galaxy survey is an order-of-magnitude increase over SDSS, Figure 9 makes it clear that the DESI bright galaxy survey will dramatically improve our knowledge of which galaxies reside in the local universe.