skip to main content

Title: SNIascore: Deep Learning Classification of Low-Resolution Supernova Spectra
We present SNIascore, a deep-learning based method for spectroscopic classification of thermonuclear supernovae (SNe Ia) based on very low-resolution (R ∼100) data. The goal of SNIascore is fully automated classification of SNe Ia with a very low false-positive rate (FPR) so that human intervention can be greatly reduced in large-scale SN classification efforts, such as that undertaken by the public Zwicky Transient Facility (ZTF) Bright Transient Survey (BTS). We utilize a recurrent neural network (RNN) architecture with a combination of bidirectional long short-term memory and gated recurrent unit layers. SNIascore achieves a <0.6% FPR while classifying up to 90% of the low-resolution SN Ia spectra obtained by the BTS. SNIascore simultaneously performs binary classification and predicts the redshifts of secure SNe Ia via regression (with a typical uncertainty of <0.005 in the range from z=0.01 to z=0.12). For the magnitude-limited ZTF BTS survey (≈70% SNe Ia), deploying SNIascore reduces the amount of spectra in need of human classification or confirmation by ≈60%. Furthermore, SNIascore allows SN Ia classifications to be automatically announced in real-time to the public immediately following a finished observation during the night.
; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ; ;
Award ID(s):
1440341 2034437
Publication Date:
Journal Name:
Sponsoring Org:
National Science Foundation
More Like this
  1. Using Zwicky Transient Facility (ZTF) observations, we identify a pair of "sibling" Type Ia supernovae (SNe Ia), i.e., hosted by the same galaxy at z = 0.0541. They exploded within 200 days from each other at a separation of 0.6″ corresponding to a projected distance of only 0.6 kpc. Performing SALT2 light curve fits to the gri ZTF photometry, we show that for these equally distant "standardizable candles", there is a difference of 2 magnitudes in their rest frame B-band peaks, and the fainter SN has a significantly red SALT2 colour c=0.57± 0.04, while the stretch values x1 of themore »two SNe are similar, suggesting that the fainter SN is attenuated by dust in the interstellar medium of the host galaxy. We use these measurements to infer the SALT2 colour standardization parameter, β = 3.5 ± 0.3, independent of the underlying cosmology and Malmquist bias. Assuming the colour excess is entirely due to dust, the result differs by 2σ from the average Milky-Way total-to-selective extinction ratio, but is in good agreement with the colour-brightness corrections empirically derived from the most recent SN Ia Hubble-Lemaitre diagram fits. Thus we suggest that SN "siblings", which will increasingly be discovered in the coming years, can be used to probe the validity of the colour and lightcurve shape corrections using in SN Ia cosmology while avoiding important systematic effects in their inference from global multi-parameter fits to inhomogeneous data-sets, and also help constrain the role of interstellar dust in SN Ia cosmology.« less
  2. ABSTRACT The 5-yr Dark Energy Survey Supernova Programme (DES-SN) is one of the largest and deepest transient surveys to date in terms of volume and number of supernovae. Identifying and characterizing the host galaxies of transients plays a key role in their classification, the study of their formation mechanisms, and the cosmological analyses. To derive accurate host galaxy properties, we create depth-optimized coadds using single-epoch DES-SN images that are selected based on sky and atmospheric conditions. For each of the five DES-SN seasons, a separate coadd is made from the other four seasons such that each SN has a correspondingmore »deep coadd with no contaminating SN emission. The coadds reach limiting magnitudes of order ∼27 in g band, and have a much smaller magnitude uncertainty than the previous DES-SN host templates, particularly for faint objects. We present the resulting multiband photometry of host galaxies for samples of spectroscopically confirmed type Ia (SNe Ia), core-collapse (CCSNe), and superluminous (SLSNe) as well as rapidly evolving transients (RETs) discovered by DES-SN. We derive host galaxy stellar masses and probabilistically compare stellar-mass distributions to samples from other surveys. We find that the DES spectroscopically confirmed sample of SNe Ia selects preferentially fewer high-mass hosts at high-redshift compared to other surveys, while at low redshift the distributions are consistent. DES CCSNe and SLSNe hosts are similar to other samples, while RET hosts are unlike the hosts of any other transients, although these differences have not been disentangled from selection effects.« less
  3. In order to identify the sources of the observed diffuse high-energy neutrino flux, it is crucial to discover their electromagnetic counterparts. To increase the sensitivity of detecting counterparts of transient or variable sources by telescopes with a limited field of view, IceCube began releasing alerts for single high-energy ( E ν  >  60 TeV) neutrino detections with sky localisation regions of order 1° radius in 2016. We used Pan-STARRS1 to follow-up five of these alerts during 2016–2017 to search for any optical transients that may be related to the neutrinos. Typically 10–20 faint ( m i P1  ≲ 22.5 mag) extragalacticmore »transients are found within the Pan-STARRS1 footprints and are generally consistent with being unrelated field supernovae (SNe) and AGN. We looked for unusual properties of the detected transients, such as temporal coincidence of explosion epoch with the IceCube timestamp, or other peculiar light curve and physical properties. We found only one transient that had properties worthy of a specific follow-up. In the Pan-STARRS1 imaging for IceCube-160427A (probability to be of astrophysical origin of ∼50%), we found a SN PS16cgx, located at 10.0′ from the nominal IceCube direction. Spectroscopic observations of PS16cgx showed that it was an H-poor SN at redshift z  = 0.2895 ± 0.0001. The spectra and light curve resemble some high-energy Type Ic SNe, raising the possibility of a jet driven SN with an explosion epoch temporally coincident with the neutrino detection. However, distinguishing Type Ia and Type Ic SNe at this redshift is notoriously difficult. Based on all available data we conclude that the transient is more likely to be a Type Ia with relatively weak Si  II absorption and a fairly normal rest-frame r -band light curve. If, as predicted, there is no high-energy neutrino emission from Type Ia SNe, then PS16cgx must be a random coincidence, and unrelated to the IceCube-160427A. We find no other plausible optical transient for any of the five IceCube events observed down to a 5 σ limiting magnitude of m i P1  ≈ 22 mag, between 1 day and 25 days after detection.« less
  4. We present optical photometry and spectroscopy of the Type II supernova ASASSN-14jb, together with Very Large Telescope (VLT) Multi Unit Spectroscopic Explorer (MUSE) integral field observations of its host galaxy and a nebular-phase spectrum. This supernova, in the nearby galaxy ESO 467-G051 ( z  = 0.006), was discovered and followed-up by the all-sky automated survey for supernovae (ASAS-SN). We obtained well-sampled las cumbres network (LCOGTN) B V g r i and Swift w 2 m 1 w 1 u b v optical, near-UV/optical light curves, and several optical spectra in the early photospheric phases. The transient ASASSN-14jb exploded ∼2 kpc abovemore »the star-forming disk of ESO 467-G051, an edge-on disk galaxy. The large projected distance from the disk of the supernova position and the non-detection of any H II region in a 1.4 kpc radius in projection are in conflict with the standard environment of core-collapse supernova progenitors and suggests the possible scenario that the progenitor received a kick in a binary interaction. We present analysis of the optical light curves and spectra, from which we derived a distance of 25 ± 2 Mpc using state-of-the-art empirical methods for Type II SNe, physical properties of the SN explosion ( 56 Ni mass, explosion energy, and ejected mass), and properties of the progenitor; namely the progenitor radius, mass, and metallicity. Our analysis yields a 56 Ni mass of 0.0210  ±  0.0025  M ⊙ , an explosion energy of ≈0.25 × 10 51 ergs, and an ejected mass of ≈6  M ⊙ . We also constrained the progenitor radius to be R *  = 580  ±  28  R ⊙ which seems to be consistent with the sub-Solar metallicity of 0.3  ±  0.1  Z ⊙ derived from the supernova Fe II λ 5018 line. The nebular spectrum constrains strongly the progenitor mass to be in the range 10–12 M ⊙ . From the Spitzer data archive we detect ASASSN-14jb ≈330 days past explosion and we derived a total dust mass of 10 −4   M ⊙ from the 3.6 μ m and 4.5 μ m photometry. Using the F U V , N U V , B V g r i , K s , 3.6 μ m, and 4.5 μ m total magnitudes for the host galaxy, we fit stellar population synthesis models, which give an estimate of M *  ≈ 1 × 10 9   M ⊙ , an age of 3.2 Gyr, and a SFR ≈0.07  M ⊙ yr −1 . We also discuss the low oxygen abundance of the host galaxy derived from the MUSE data, having an average of 12 + log(O/H) = 8.27 +0.16 −0.20 using the O 3 N 2 diagnostic with strong line methods. We compared it with the supernova spectra, which is also consistent with a sub-Solar metallicity progenitor. Following recent observations of extraplanar H II regions in nearby edge-on galaxies, we derived the metallicity offset from the disk, being positive, but consistent with zero at 2 σ , suggesting enrichment from disk outflows. We finally discuss the possible scenarios for the unusual environment for ASASSN-14jb and conclude that either the in-situ star formation or runaway scenario would imply a low-mass progenitor, agreeing with our estimate from the supernova nebular spectrum. Regardless of the true origin of ASASSN-14jb, we show that the detailed study of the environment roughly agree with the stronger constraints from the observation of the transient.« less
  5. We present 42 rapidly evolving (time spent above half-maximum brightness t1/2<12d) extragalactic transients from Phase I of the Zwicky Transient Facility (ZTF), of which 22 have spectroscopic classifications. This is one of the largest systematically selected samples of day-timescale transients, and the first with spectroscopic classifications. Most can be classified as core-collapse supernovae (SNe), and we identify several predominant subtypes: (1) subluminous Type IIb or Type Ib SNe; (2) luminous Type Ibn or hybrid IIn/Ibn SNe; and (3) radio-loud, short-duration luminous events similar to AT2018cow. We conclude that rates quoted in the literature for rapidly evolving extragalactic transients are dominatedmore »by the subluminous events (mostly Type IIb SNe). From our spectroscopic classifications and radio, X-ray, and millimeter-band upper limits, we are motivated to consider the AT2018cow-like objects a distinct class, and use ZTF's systematic classification experiments to calculate that their rate does not exceed 0.1% of the local core-collapse SN rate, in agreement with previous work. By contrast, most other events are simply the extreme of a continuum of established SN types extending to ordinary timescales. The light curves of our objects are very similar to those of unclassified events in the literature, illustrating how spectroscopically classified samples of low-redshift objects in shallow surveys like ZTF can be used to photometrically classify larger numbers of events at higher redshift.« less