Discovery

The intermediate Palomar Transient Factory (iPTF) first detected iPTF14hls on 2014 September 22.53 (Extended Data Fig. 1) using the iPTF real-time image-subtraction pipeline23. No source was seen at that position when it was previously visited by iPTF and by the All Sky Automated Survey for Supernovae (ASAS-SN)24 on 2014 May 6.19 and 2014 May 20–28 down to 3σ limiting magnitudes of R < 20.95 and V < 18.7, respectively. The source was observed by iPTF again on 2014 October 13, October 31, November 4 and November 10 before being saved and given a name as part of routine iPTF transient scanning. On 2014 November 18, iPTF14hls was independently discovered by the Catalina Real-Time Transient Survey25 as CSS141118:092034+504148, and later the event was reported to the Transient Name Server as AT 2016bse and Gaia16aog. On 2015 February 3, upon routine Las Cumbres Observatory (LCO) rescanning of previously saved iPTF candidates, we noticed the peculiar decline and subsequent rise of the light curve, and began an extensive campaign of spectroscopic and multi-band photometric follow-up observations.

Follow-up imaging

Follow-up imaging was obtained with the Palomar 48-inch Oschin Schmidt telescope (P48), the Palomar 60-inch telescope (P60)26 using both the GRBCam and the SEDM instruments, the LCO27 network 1-m and 2-m telescopes, and the 0.8-m Tsinghua University-NAOC telescope (TNT)28 at the Xinglong Observatory. The TNT photometry is presented, together with Catalina Sky Survey (CSS; http://nesssi.cacr.caltech.edu/catalina/20141118/1411181490344143272.html) and Gaia (http://gsaweb.ast.cam.ac.uk/alerts/alert/Gaia16aog/) photometry downloaded from their respective websites, in Extended Data Figure 2. P48 images were first pre-processed by the Infrared Processing and Analysis Center (IPAC)29. Image subtraction and point-spread-function fitting were then performed30 using pre-explosion images as templates. Magnitudes were calibrated to observations of the same field by the Sloan Digital Sky Survey (SDSS) DR1031. P60 images were pre-processed using a PyRAF-based pipeline26. Image subtraction, photometry extraction and calibration were performed with the FPipe pipeline32 using SDSS images as references. LCO images were pre-processed using the Observatory Reduction and Acquisition Control Data Reduction pipeline (ORAC-DR)33 up to 2016 May 4, and using the custom Python-based BANZAI pipeline afterward. Photometry was then extracted using the PyRAF-based LCOGTSNPipe pipeline34 to perform point-spread-function fitting and calibration to the AAVSO Photometric All-Sky Survey35 for BV-band data and SDSS DR836 for gri-band data. TNT images were reduced with standard Image Reduction and Analysis Facility (IRAF) routines; point-spread-function fitting was performed using the SNOoPy package and calibrated to the SDSS DR937 transformed to the Johnson system38. We correct all photometry for Milky Way extinction39 extracted via the NASA Extragalactic Database. Pre-explosion nondetection limits are presented in Extended Data Figure 3.

We fitted a blackbody spectral energy distribution to every epoch of LCO photometry containing at least three of the BVgi filters obtained within 0.4 days of each other (we exclude r-band and R-band data from the fits owing to contamination from the Hα line). For each epoch we perform a blackbody fit using Markov Chain Monte Carlo simulations through the Python emcee package40 to estimate the blackbody temperature and radius at the measured distance to iPTF14hls of 156 Mpc.

Follow-up spectroscopy

Spectra of iPTF14hls were obtained with the Floyds instrument mounted on the northern LCO 2-m telescope27, the Andalucia Faint Object Spectrograph and Camera (ALFOSC) mounted on the 2.5-m Nordic Optical Telescope (NOT), the Device Optimized for the LOw RESolution (DOLoRes) mounted on the 3.6-m Telescopio Nazionale Galileo (TNG), the Low Resolution Imaging Spectrometer (LRIS)41 mounted on the Keck-I 10-m telescope, the DEep Imaging Multi-Object Spectrograph (DEIMOS)42 mounted on the Keck-II 10-m telescope, the Double Beam Spectrograph (DBSP)43 mounted on the Palomar 200-inch telescope (P200), the Beijing Faint Object Spectrograph and Camera (BFOSC) on the Xinglong 2.16-m telescope of the National Astronomical Observatories of China, the Yunnan Faint Object Spectrograph and Camera (YFOSC) on the Lijiang 2.4-m telescope of the Yunnan Observatories, and the DeVeny spectrograph mounted on the 4.3-m Discovery Channel Telescope (DCT). The Floyds spectra were reduced using the PyRAF-based floydsspec pipeline. The ALFOSC and DOLORES spectra were reduced using custom MATLAB pipelines. The LRIS spectra were reduced using the IDL LPipe pipeline. The DEIMOS spectrum was reduced using a modified version of the DEEP2 pipeline44,45 combined with standard PyRAF and IDL routines for trace extraction, flux calibration and telluric correction. The DBSP spectrum was reduced using custom IRAF and IDL routines. The BFOSC, YFOSC and DeVeny spectra were reduced using standard IRAF procedures. No Na i D line absorption is seen at the redshift of the host galaxy, indicating very low host-galaxy extinction at the supernova position.

We fitted each iPTF14hls spectrum to a library of type II supernovae (which includes a full set of SN 1999em spectra22) using Superfit47. We then calculate the average best-fitting supernova phase, weighing all the possible fits by their corresponding fit scores. We repeat this process for cutouts of the iPTF14hls spectra centred around the Hα, Hβ and Fe ii λ = 5,169 Å features (separately). The weighted-average best-fit phases for each cutout are presented in Extended Data Figure 4. iPTF14hls can be seen to evolve more slowly than other type II supernovae by a factor of approximately 10 when considering the entire spectrum, as well as when considering the Hβ and the Fe ii λ = 5,169 Å features separately, and by a factor of 6–7 when considering the Hα emission feature separately.

Expansion velocities for different elements in iPTF14hls were measured by fitting a parabola around the minimum of the absorption feature of their respective P Cygni profiles. The difference between the minimum of the best-fit parabola and the rest wavelength of the line was translated to an expansion velocity. The endpoints of each parabolic fit were chosen manually for each line, so that they would remain the same for all spectra. Uncertainties in the velocities were estimated by randomly varying these endpoints 1,000 times by ±5 Å around their original values.

iPTF14hls is probably not powered by interaction

As mentioned in the main text, interaction between supernova ejecta and pre-existing, dense circumstellar material could cause an increase in luminosity. However, iPTF14hls does not display the spectral line profiles typically seen in such cases (Extended Data Fig. 5).

In some interaction models the collision of the supernova ejecta and the circumstellar material occurs outside the broad-line-forming region, diluting the line emission. Focusing on the approximately 50% luminosity increase of iPTF14hls between rest-frame day 207 and day 232 after discovery (Fig. 1), we find that the spectra taken on day 207 and day 232 are identical up to a global normalization factor. This indicates that the increase in luminosity is equal at all wavelengths, in contrast to the expected line dilution from interaction (Extended Data Fig. 6).

Additional possible indicators of interaction are strong X-ray and/or radio emission. We observed the location of iPTF14hls with the X-Ray Telescope (XRT)48 onboard the Swift satellite49 on 2015 May 23.05. A total of 4.9 ks of live exposure time was obtained on the source. We use online analysis tools50,51 to search for X-ray emission at the location of iPTF14hls. No source is detected with an upper limit on the 0.3–10.0-keV count rate of <2.3 × 10−3 counts s−1. Assuming a power-law spectrum with a photon index of Γ = 2 and a Galactic H column density52 of 1.4 × 1020 cm−2, this corresponds to an upper limit on the unabsorbed 0.3–10.0-keV flux of f X < 8.4 × 10−14 erg cm−2 s−1. At the luminosity distance of iPTF14hls this corresponds to a luminosity limit of L X < 2.5 × 1041 erg s−1 (which is roughly 10−2 of the peak bolometric luminosity). The lack of X-ray emission disfavours strong interaction in iPTF14hls, though some interacting supernovae display X-ray emission fainter than the limit we deduce here53.

We observed iPTF14hls also with the Arcminute Microkelvin Imager Large Array (AMI-LA)54 at 15 GHz on 2015 May 18.59, May 19.77, May 23.63, May 25.65, May 28.66, and May 31.62. Quasars 3C48 and J2035+1056 were used as the flux/bandpass and phase calibrators, respectively. Radio frequency interference excision and calibration of the raw data were done with a fully automated pipeline AMI-REDUCE55,56. The calibrated data for the supernova were imported into CASA (Common Astronomy Software Applications) and imaged independently for each epoch into 512 × 512 pixel maps (4″ per pixel) using the clean task. A similar imaging scheme was used for the concatenated data from all the epochs as well. The supernova was not detected on any of the individual epochs, with 3σ upper limits of 60–120 μJy. The combined 3σ upper limit is 36 μJy. There is a 5%–10% absolute flux calibration uncertainty that we have not considered in these upper limits. On 2016 June 10, iPTF14hls was observed with the VLA at 6.1 GHz. The VLA data were reduced using standard CASA software routines where J0920+4441 and 3C286 were used as phase and flux calibrators. No radio emission was observed at the supernova position to a 3σ upper limit of 21.3 μJy. At the luminosity distance of iPTF14hls, this corresponds to 6.2 × 1026 erg s−1 Hz−1, which is fainter than the radio emission of most interacting supernovae53. We conclude that iPTF14hls does not show any of the signatures normally seen in supernovae powered by interaction.

A possible central-engine power source for iPTF14hls

A central engine such as the spindown of a magnetar12,57,58 or fallback accretion onto a black hole13,59 created after core collapse (assuming the material falling back has sufficient angular momentum to form a disk) could inject power to the supernova, although (as noted in the main text) this may fail to reproduce the observed iron and hydrogen line velocity difference. A magnetar (with an initial spin period of about 5–10 ms and a magnetic field of about (0.5–1) × 1014 G) could produce the observed average luminosity and timescale of iPTF14hls12. However, the analytical magnetar light curve required to fit the late-time decline overpredicts the early-time emission of iPTF14hls (Extended Data Fig. 2) and produces a smooth rather than variable light curve12,13.

For a black-hole central engine, on the other hand, instabilities in the accretion flow might produce strong light-curve variability, as seen in active galactic nuclei20. In this case, the light curve is expected to eventually settle onto a t−5/3 decline rate21 after the last instability. Such a decline rate is indeed observed for iPTF14hls starting around day 450 (Extended Data Fig. 2), supporting a black-hole power source. We conclude that iPTF14hls does not show the expected signatures of magnetar power (using available analytical models), but might be consistent with black-hole accretion power.

No signs of asymmetry in iPTF14hls

A possible explanation for the high luminosities and apparent emitted energy of iPTF14hls, as well as the discrepancy between its line-forming versus blackbody radii, is strong asymmetry in the explosion. Such asymmetry would be indicated by a polarization signal.

We observed iPTF14hls with the Andalucia Faint Object Spectrograph and Camera (ALFOSC) mounted on the 2.5-m Nordic Optical Telescope (NOT) in polarimetric mode on 2015 November 03 in the R band, and on December 15 in the V band (we also obtained observations on 2015 October 28 and November 14 but we discard them because of very poor observing conditions). We used a 1/2-wave plate in the FAPOL unit (the polarimeter unit in the Filter And Shutter Unit, FASU) and a calcite plate mounted in the aperture wheel, and observed at four different retarder angles (0°, 22.5°, 45°, 67.5°). The data were reduced in a standard manner, using bias frames and flat fields without the polarization units in the light path. The field of view contains one bright star that can be used for calibration and for determining the interstellar polarization in the Galaxy. The low Galactic extinction towards iPTF14hls implies an expected interstellar polarization value60 of <0.13%.

To measure the fluxes we performed aperture photometry, and to compute the polarisation we followed standard procedures61. For our epoch with the best signal-to-noise (2015 November 03), we measure P = 0.40% ± 0.27% for iPTF14hls and P = 0.17% ± 0.09% for the comparison star, in agreement with the interstellar polarization prediction. These results suggest that iPTF14hls is close to spherically symmetric, similar to what is observed for type II-P supernovae during their plateau phase62. The 2015 December 15 epoch yields a lower precision (P = 1.1% ± 0.7% for iPTF14hls and P = 0.80% ± 0.23% for the comparison star), but is still consistent with very low asphericity.

Expansion velocities of iPTF14hls

In a supernova, the ejecta are in homologous expansion—that is, the radius of the ejecta at time t evolves as r = vt, with faster material at larger radii. Even for perfectly mixed ejecta, at any given time, spectral lines of different elements form in different regions. Specifically, the iron (Fe) lines are formed at smaller radii than the hydrogen (H) lines and therefore display a lower velocity. This is also the case in iPTF14hls. As time passes and the ejecta expand and recombine, the line-forming region of each element moves inward in mass to a region where the outflow is slower. This is why, normally, the velocity of all lines is observed to decrease with time. Thus, following the line velocity over a wide range of time (and hence mass coordinates) provides a ‘scan’ of the velocity profile over a large range of the ejecta. Although different lines are formed in different regions, all line-forming regions scan the velocity profile of the same ejecta. Therefore, if there is a large velocity gradient in the ejecta, we expect to see both a large velocity difference between the Fe and H lines as well as considerable evolution in the velocity of each line as the material expands. These two features are seen clearly in the typical case of SN 1999em (Extended Data Fig. 7). However, this is not the case in iPTF14hls. On the one hand, there is a large difference between the H and Fe line velocities, indicating a large velocity gradient in the ejecta. On the other hand, the velocity of each line shows almost no evolution in time between day 100 and day 600 after discovery. If the line-forming material were ejected at the time of discovery, then this time span corresponds to a change by a factor of about 6 in radius. In this case, the lack of observed velocity evolution indicates a very shallow velocity gradient in the ejecta, which is inconsistent with the large velocity difference between the lines. However, if the ejection of the line-forming material took place before discovery, then the relative change in radius during the observations is small, indicating that the position of the line-forming region does not change much, potentially solving the apparent contradiction.

The line-forming region of iPTF14hls

The nearly constant line velocities measured in iPTF14hls suggest that the lines form in a massive shell, perhaps ejected prior to the explosion. Here we estimate the mass and energetics required for such a shell to produce the observed line features.

Consider a uniform shell of mass M with a radius r and width Δr. The number density of hydrogen atoms in the shell is: where Y H ≈ 0.9 is the number fraction of hydrogen and μ ≈ 1.34 is the mean atomic mass for solar gas (m p is the proton mass). In a rapidly expanding, homologous outflow ejected at a time t ej , the strength of a spectral line is characterized by the Sobolev optical depth approximation: where n l is the number density of atoms in the lower level, f is the line oscillation strength, t ej is the time since explosion, m e is the mass of the electron and λ 0 is the line rest wavelength. For a line to produce a noticeable absorption component in the spectra, it must have τ Sob ≳ 1.

To estimate the populations in the lower level of the line transition (for the Balmer series this is the n = 2 level), we apply the nebular approximation63, which assumes that the mean intensity of the radiation field at a radius above a nearly blackbody photosphere is J ν (r) = W(r)B ν (T bb ), where B ν is the Planck function, T bb is the temperature of the photosphere, and W(r) is the geometrical dilution factor of the radiation field: Here, r p is the photospheric radius and the last expression assumes . For a two-level atom subject to this radiation field, the number density in the n = 2 excited state is: where (n 1 , n 2 ) and (g 1 , g 2 ) are, respectively, the number density and statistical weights of the n = 1 and n = 2 levels, and ΔE 1,2 is the energy difference between the levels.

Since essentially all of the hydrogen in the shell will be neutral and in the ground state, n 1 ≈ n H . The Sobolev optical depth is then:

Using g 1 = 2, g 2 = 8, ΔE 1,2 = 10.2 eV, λ 0 = 6,563 Å (for the Hα transition), and f = 0.64, and taking T = 6,500 K, Δr = Δvt ej , and r = vt ej gives (where M ⊙ is the solar mass). Though approximate, this argument demonstrates that a shell with a mass of the order of a few tens of solar masses is probably required for producing Balmer absorption lines throughout the approximately 600-day duration of the iPTF14hls light curve. The corresponding kinetic energy of the outburst is about 1052 erg. In the case where the shell was ejected before the first iPTF14hls observations, the mass and energy required would increase. However, the mass required to associate the line-forming region with the 1954 eruption would be about 107M ⊙ , and hence not reasonable, implying that the line-forming region was ejected in a separate, more recent, eruption.

For comparison, the electron-scattering optical depth of the shell is: where σ T is the Thomson cross-section and x H II is the fraction of ionized hydrogen. The shell will be largely neutral , because the region where the radiation field is sufficient to ionize hydrogen occurs at the photosphere (r p ) where the recombination front forms. The shell radius is much larger than r p , so the radiation field is strongly diluted. Thus, while the shell can form line features, it will be optically thin in the continuum and allow most of the pseudo-blackbody continuum from the photosphere to pass through.

The velocity of 6,000 km s−1 seen for Hα at day 600 after discovery is also seen for Hβ at day 200 after discovery. If we calculate the optical depth (see equation (5)) for Hβ (plugging in the parameters for day 200 + t 0 , where t 0 is the offset between the ejection of the shell and discovery) and equate it to that of Hα at day 600 + t 0 , then we can solve for the ejection time t 0 , assuming that the optical depths for Hα and Hβ were the same when each was observed at 6,000 km s−1, and that the entire shell was ejected simultaneously. Using λ 0 = 4,861 Å and f = 0.12 for the Hβ transition, we find t 0 ≈ 100–200 days (the main source of error is the uncertainty in the precise temperature difference between the two epochs), meaning that the line-forming shell was ejected 100–200 days before discovery. We have deep nondetection limits for part of this epoch (Extended Data Fig. 3), suggesting that the ejection of the shell could have been a low-luminosity event. This estimation of the ejection time, however, relies on many simplifying assumptions, so it should be considered only as an approximation.

A historical outburst at the position of iPTF14hls

The Palomar Observatory Sky Survey (POSS)64 observed the field of iPTF14hls on 1954 February 23 in the blue and red filters. POSS-II65 then re-observed the field on 1993 January 2 in the blue filter and on 1995 March 30 in the red filter. We obtained these images through the STScI Digitized Sky Survey and we find a source at the position of iPTF14hls in the blue image from POSS that is not present in the blue image from POSS-II (Extended Data Fig. 8). We do not see this source in either of the red images, but they are not as deep as the blue images (the limiting magnitude is roughly 20 for the red images compared to 21.1 for the blue images)64.

We register the POSS blue image to the POSS-II blue image using the IRAF task wregister. We then use the apphot package in PyRAF, with a 3-pixel aperture, to measure the flux in six stars in the field near the position of iPTF14hls to determine a zero-point offset for the two images. We find an offset of 0.132 ± 0.050 mag. We then perform the same measurement around the nucleus of the host galaxy of iPTF14hls and find an offset of 0.141 mag, consistent with the zero-point offset. Next we perform the same aperture photometry measurement at the position of iPTF14hls in both images. We find a magnitude difference of 0.31 ± 0.14 over the host-galaxy level confirming the presence of an outburst in the 1954 image at the position of iPTF14hls at a 2.2σ confidence level. Owing to the nonlinear nature of the photographic plates used in the two POSS surveys, as well as differences between the filters65, we cannot perform meaningful image subtraction between the POSS epochs to obtain more accurate photometric measurements. We consider this confidence level to be a conservative estimate; the outburst can be seen clearly by eye in the images (Extended Data Fig. 8).

We calibrate the six stars used for the zero-point comparison to SDSS u-band plus g-band fluxes (the POSS blue filter roughly covers the SDSS u and g bands)64 and find that the magnitude of the 1954 outburst (after removing the host-galaxy contribution) is 20.4 ± 0.1(stat) ± 0.8(sys). The first error is statistical and due to photometric measurement uncertainties, while the second error is systematic and caused by the calibration to SDSS (the large error value is probably produced by filter and detector differences between POSS and SDSS).

This corresponds to an absolute magnitude for the outburst of approximately −15.6 at the luminosity distance of iPTF14hls (this is only a lower limit on the peak luminosity of the eruption, as we have only one epoch of observations). Such an eruption may be produced by the pulsational pair instability2,3,4,5. Eruptions of similar luminosity (though probably caused by different instabilities) are inferred to be common in type IIn supernova progenitors in the last year prior to explosion66. Spectra and broad-band colours are available for three such possible outbursts—a precursor to PTF10bjb66, PTF13efv (a precursor to SNHunt275)67 and the first 2012 outburst of SN2009ip68—all of which display rather flat continuum emission, consistent with the limited colour information we have for the 1954 outburst of iPTF14hls (that is, the red nondetection limit being about 0.4 magnitudes brighter than the blue detection).

Given the host galaxy size of about 10–100 times the centroiding error of the outburst, and a typical supernova rate of about 1/100 per galaxy per year, there is a probability of a few per cent that the detected outburst is an unrelated supernova that happened to occur at the position of iPTF14hls.

The rate of iPTF14hls-like events

On 2014 November 18, iPTF14hls was independently discovered by the Catalina Real-Time Transient Survey25 as CSS141118:092034+504148, and more recently the event was reported to the Transient Name Server as AT 2016bse and Gaia16aog. The fact that it was discovered multiple times, but dismissed as a run-of-the-mill type II-P supernova, suggests that similar events may have been missed in the past. We ourselves would not have noticed the unique properties of iPTF14hls had the iPTF survey scheduler not automatically continued to monitor the position of iPTF14hls. In addition, if iPTF14hls-like events are limited to low-mass galaxies, then targeted transient surveys would have missed them completely.

To our knowledge, iPTF14hls is the only supernova ever discovered to show such long-lived, slowly evolving type II-P-like emission. The PTF and iPTF surveys discovered 631 type II supernovae, indicating that iPTF14hls-like events could be about 10−3–10−2 of the type II supernova rate. Since luminous, long-lived varying events could be easier to detect in transient surveys compared to normal supernovae, the true volumetric rate of iPTF14hls-like events could be much lower. On the other hand, we cannot rule out whether such events were discovered in the past but dismissed as normal type II-P supernovae after one spectrum with no subsequent follow-up observations or as possible active galactic nuclei owing to the light curve behaviour. It is therefore not possible to calculate a precise rate for iPTF14hls-like events based on this single discovery, but whatever the explosion channel, it is likely to be rare. Even so, the Large Synoptic Survey Telescope could find hundreds of iPTF14hls-like events in its decade-long survey of the transient sky (and more if iPTF14hls-like events are more common in the early Universe, as is indicated by the possible low-metallicity environment of iPTF14hls).

The host galaxy of iPTF14hls

We obtained a spectrum of the host galaxy of iPTF14hls on 2015 December 11 with the Low Resolution Imaging Spectrometer (LRIS)41 mounted on the Keck-I 10-m telescope. The spectrum was reduced using standard techniques optimized for Keck+LRIS by the CarPy package in PyRAF, and flux calibrated with spectrophotometric standard stars obtained on the night of our observations in the same instrument configuration. The host-galaxy spectrum, which is available for download via WISeREP46, shows clear detections of Hα, Hβ, [O ii] λ = 3,727 Å and [O iii] λ = 4,958 Å and 5,007 Å, which we use to determine a redshift of 0.0344. A faint detection of [N ii] λ = 6,583 Å is also possible, but the feature is difficult to confirm because the continuum is contaminated by broad Hα emission from the nearby supernova. All of the lines are weak (equivalent width <20 Å) and no other lines are strongly detected. We extracted the fluxes of all lines by fitting Gaussians to their profiles (Extended Data Table 1), and calculated the metallicity by fitting69 the line-strength ratios using several different diagnostics and calibrations (Extended Data Table 2). We find a range of metallicity estimates of 12 + log(O/H) = 8.3–8.6, corresponding to about (0.4–0.9)Z ⊙ (where Z ⊙ is the solar metallicity)70. A low metallicity could help explain how the progenitor of iPTF14hls retained a very massive hydrogen envelope. Future, more direct environment studies will be able to better probe the metallicity at the explosion site.

We fitted the SDSS ugriz photometry of the host galaxy71 with standard spectral energy distribution fitting techniques72 using the BC0373 stellar population synthesis models. Assuming a metallicity of 0.5Z ⊙ , the best-fit total stellar mass is (3.2 ± 0.5) × 108M ⊙ , similar to that of the Small Magellanic Cloud.

Data availability

The photometric data that support the findings of this study are available in the Open Supernova Catalog81, https://sne.space/sne/iPTF14hls/. The spectroscopic data that support the findings of this study are available on the Weizmann Interactive Supernova data REPository (WISeREP)46, https://wiserep.weizmann.ac.il/, and on the Open Supernova Catalog. Source Data for Figs 1, 3 and 4, and for Extended Data Figures 2, 3 and 4 are provided with the online version of the paper.