Delayed Circumstellar Interaction for Type Ia SN 2015cp Revealed by an HST Ultraviolet Imaging Survey

To generate an initial list of potential targets, we used discoveries from both professional and amateur surveys, such as the original and intermediate Palomar Transient Factory (PTF and iPTF; Rau et al. 2009), the La Silla Quest survey (LSQ; Baltay et al. 2013), the All-Sky Automated Survey for Supernovae (ASAS-SN¹³ ), the Panoramic Survey Telescope and Rapid Response System (PanSTARRS; Chambers et al. 2016), the Optical Gravitational Lensing Experiment (OGLE; Udalski et al. 2015), the Lick Observatory Supernova Search (LOSS) with the Katzman Automatic Imaging Telescope (KAIT; Filippenko et al. 2001), the Master Global Robotic Network (Lipunov et al. 2010), and SNHunt which is part of the Catalina Real Time Survey (CRTS; Drake et al. 2009). We also used the private PTF/iPTF Marshal, the private Las Cumbres Observatory’s Supernova Exchange database, the public Astronomer’s Telegrams, the public Transient Name Server, the public supernova database WISeREP (Yaron & Gal-Yam 2012), and the public Open Supernova Catalog (Guillochon et al. 2017) to form an initial list of possible targets.

Our goal in choosing targets was to maximize our chances of a detection, so we prioritized SNe Ia that were reported to exhibit characteristics known or suspected to correlate with the single-degenerate scenario and/or the presence of CSM. One of the characteristics of SNe Ia that is known to correlate with such qualities is an overluminous light curve and/or a spectrum similar to those of SN Ia 1991T (Fisher & Jumper 2015; Leloudas et al. 2015; Katsuda et al. 2015), a subtype that occurs more often in relatively young stellar populations (e.g., Howell et al. 2009). SNe Ia with a total-to-selective extinction (R_V) lower than Galactic values exhibit higher photospheric velocities than normal, and these SNe Ia are more often associated with a relatively young, metal-rich population (Wang et al. 2013), suggesting (indirectly) that they may have a CSM-producing progenitor (i.e., assuming the CSM would have a different value of R_V than Milky Way dust). For these reasons, we have prioritized SNe Ia that exhibit a “91T-like” light curve or spectrum, high photospheric velocities (HV), blueshifted Na i D absorption (see Section 1), and/or hosts with younger stellar populations. As a control set, we also included a small fraction of SNe that exhibited strong signatures of CSM interaction around their time of discovery (SNe Ia–CSM and SNe IIn).

Our final sample, listed in Table 1, includes 39 SNe Ia with a (tentative, at least) classification of SN 1991T-like (i.e., a spectroscopic match or an overluminous peak brightness), 16 SNe Ia classified as normal, and 9 SNe IIn and/or SNe Ia–CSM; the remaining 16 are SNe Ia of other subtypes. In column 5 of Table 1, we provide an abbreviated classification for each of our targets, and Table 2 defines those abbreviations. Column 6 of Table 1 provides the discovery and/or classification references for each of our targets, when they are publicly available; in a few cases the full classification relies on private survey data (e.g., PTF). The redshifts and distances listed in columns 3 and 4 represent the best measurements or estimates available: host-galaxy spectra lead to redshifts with >3 significant figures, whereas SN spectral fits are less precise, and most of the distance estimates are based on the redshift (assuming a flat cosmology with Ω_M = 0.3 and Ω_Λ = 0.7) except for a few nearby SNe with redshift-independent distances (e.g., 2014J). In pursuit of our goal to maximize our chances of detection, we also limited our sample to SNe within a distance of ≲350 Mpc and with phases of ∼1–3 yr post-explosion during Cycle 24, i.e., after the supernova’s emission had faded and later than most SNe Ia are routinely monitored.

In choosing targets for this program, we also considered several conditions for technical feasibility. To maximize the probability of a detection, we rejected any SNe that appeared to be in regions of high host-galaxy UV background. To ensure that in the case of any detection we could make as thorough a physical interpretation as possible, we also prioritized objects with denser spectroscopic and photometric coverage around peak brightness.

For our survey, we chose to use HST WFC3+F275W because it covers the expected emission region and is wide enough to offer a decent signal-to-noise ratio. Images with F275W will also be free from the light-leak contamination suffered by F225W. Three other advantages of an NUV search are as follows. (1) We are unlikely to mistake an SN Ia light echo for CSM interaction because near-peak SNe Ia do not exhibit much NUV flux. While SNe Ia with CSM might be more likely to produce a light echo, it would be very faint (e.g., as seen for SN 2006X; Wang et al. 2008). (2) Late-time SNe Ia without CSM interaction are very faint in the NUV and will not contaminate our results. (3) Although a shocked companion star might cause excess UV flux, it would be below the detection threshold for most of our targets (Pan et al. 2012; Graham et al. 2015b).

As this was the first late-time NUV imaging program attempted for SNe Ia, and any data at all would constrain the amount of CSM, we designed our program to be able to observe a large number of SNe Ia. HST Snapshot observations with durations in the 20–30 minute range have the highest completion rate, and a 20 minute exposure time would be sufficient for our science goals, because it would yield limiting magnitudes of m_F275W ≳ 25 mag (AB). In order to minimize overheads and maximize integration time, we kept individual exposures >348 s to avoid the 5.8 minute buffer dump, and to minimize the effects of cosmic rays (CRs), we obtained two frames instead of just one. After including overheads, our 20 minute Snapshot visits all comprised two individual exposures that each have an integration time of 429 s.

This HST Snapshot survey ran through Cycle 24, with observations occurring randomly in late 2016 and throughout 2017. Because Snapshot programs stay live for two cycles, there is also one observation in 2018 May (SN 2014ch). The observation dates for successful targets are listed in Table 3. Our program obtained 72 out of our 80 requested targets, for an overall completion rate of 90%. This is a very high completion rate for a Snapshot program, which typically achieves closer to 30% completeness.

We downloaded the processed data from the Mikulski Archive for Space Telescopes (MAST) and used the images that were combined with AstroDrizzle and corrected for charge transfer efficiency (the drc files). Since only two exposures are obtained per visit, the CR removal is not complete, but the CRs are still easily identified and are not a source of contamination for our survey. Our choice of relatively short exposure times and the F275W filter results in quite sparsely populated images. As described below, we find that the original SN coordinates from their discovery surveys, combined with the inherent astrometric solution of the AstroDrizzled images, are sufficient to confirm that 64 of our HST observations are nondetections. We find this to be the case even though these two components lead to astrometric uncertainty radii of 1″–2″, which is significantly larger than the astrometric uncertainties derived from (for example) coregistering the HST images to ground-based images that contain the SN. For the eight observations with potential detections, however, we do use images containing the SN to revise the astrometry of our HST images (described later, in Sections 3.3–3.5). We do not need to refine the astrometry of our HST images unless there is a marginal detection.

To determine the absolute astrometric uncertainty radius for the location of the SN site in the HST images, we combined the uncertainties from the original ground-based imaging that discovered the SN with those for our WFC3 images. Ground-based imaging provides absolute astrometry of ≲1″—this is typical even for the ASAS-SN project which has large pixels (Holoien et al. 2017). Usually the astrometric precision is much better, and so we use 1″ as the 3σ upper limit on the absolute astrometric uncertainty from ground-based surveys’ reported coordinates.

The absolute astrometric uncertainty of the HST WFC3 images is dominated by the guide stars’ astrometric errors, and the next largest contribution is from any failures during guiding or guiding difficulties that may cause the telescope to roll; other sources such as geometric distortion have a much smaller contribution (Deustua 2016). We use the image headers to verify that fine-lock guiding was successful and use the telescope jitter data to verify that the roll angle was stable during the observations (i.e., we use the jif tables). We obtain information about the guide stars from the headers and query the Guide Star Catalog 2.3 through the Vizier¹⁴ service to obtain their coordinate uncertainties. We find that for all images, the guide-star coordinates are the largest source of error in the HST absolute astrometry (i.e., hundreds instead of several to tens of milliarcsec), and so these errors represent the absolute astrometric precision at the 1σ level.¹⁵ For the final absolute astrometric uncertainty radius in which we search for NUV sources, we add 10 to 3 times the guide-star error (i.e., 3σ) and use this as the combined ground- and space-based absolute astrometric uncertainty. This radius is listed in Table 3 under “Total Astrometric Uncertainty.”

We use Source Extractor (Bertin & Arnouts 1996) to detect sources in our images, using the parameter values set as listed in Table 4. We list the number of detected sources within the astrometric uncertainty radius for each of our SNe in Table 3. We then visually review all images, reduced and raw frames, to classify these sources. In almost all cases, detected sources can be easily identified as a CR residual (i.e., it appeared in only one of our two original frames, or had a flux profile that deviated obviously from the HST point-spread function, PSF). The results of our visual review are reported in column 6 of Table 3. Cases that are even slightly ambiguous are treated individually in Sections 3.3–3.5, and all other observations are considered confirmed nondetections.

In order to determine the limiting magnitude of our HST images, we inject simulated WFC3 point sources with a variety of fluxes and measure the apparent magnitude at which we recover a point source 50% of the time.

Our preselection of candidates in regions of low NUV surface brightness also means that for an estimate of the point-source limiting magnitudes, we can plant fake sources in the surrounding region and not only the exact pixel location of the SN, which is helpful to minimize the amount of computational processing. In order to plant enough fakes, we do 1000 realizations, injecting 10 simulated point sources each time, with random locations within a ∼4″ box around the SN location. We use fluxes that cover a range of values, from bright enough to always be detected, to faint enough to never be detected. We use the TinyTim PSF (Krist 1993) via the TinyTim web service¹⁶ and generate the correct PSF for the WFC3 UVIS detector at the chip location for the site of the SN to ensure that the PSF distortion is accounted for. We inject directly into the post-AstroDrizzled drc images. Although this will result in slightly less accurate limiting magnitudes than injecting into the flc images, this effect is ≪0.1 mag. Injecting fakes directly into the drc images saves a significant amount of processing time and is completely sufficient for our science goals (which are achieved with a mean error in limiting magnitude of 0.07 mag, as described below).

We then run Source Extractor on each image realization, using the same parameters as in Table 4. We match the detected sources to the injected sources with a 2 pixel radius and determine the fraction of injected sources detected as a function of magnitude in bins of 0.05 mag (using the input, and not the recovered, magnitude in order to automatically incorporate the aperture correction). We then perform a spline fit to this histogram and calculate the apparent magnitude of a source with a 50% probability of detection and use that as our limiting magnitude. We convert Poisson errors for the number of objects per bin into an uncertainty on the limiting magnitude. The results are listed in column 7 of Table 3, in AB magnitudes. The limiting magnitudes of all of our images have a mean value of 25.85 mag and a standard deviation of 0.08 mag; the mean error in limiting magnitude is 0.07 mag. These limits are used to constrain the frequency of late-onset CSM interaction and the physical characteristics of CSM in SN Ia progenitor systems in Section 5.

In this section, we demonstrate our in-depth image review that confirmed the detections of SN Ia–CSM/SN IIn ASASSN-15og and SN Ia 2015cp and provide some analysis of their NUV point source.

3.3.1. ASASSN-15og

For ASASSN-15og, we find that three of the five sources within the 22 astrometric uncertainty radius can be confirmed as CR residuals, as they appear in only one of the two original flc images. We label the other two sources as objects A and B in Figure 1. As done in Section 3.4.1, the simplest first step is to try to improve the astrometric uncertainty, to see if an association with the SN’s location can be ruled out for either or both of them.

Zoom In Zoom Out Reset image size

Imaging of ASASSN-15og with the Neil Gehrels Swift Observatory is publicly available.¹⁷ We find that Swift images from the Ultraviolet and Optical Telescope (UVOT) in filters UBV obtained on 2015 August 16 (UT dates are used throughout) contain both the SN and a nearby star (∼55″ W of the SN), which is also visible in our HST F275W image. The Swift V-band image has the smallest astrometric uncertainty. We use Source Extractor to obtain the centroids of both the SN and the star in the Swift image, and the centroids of the star in our HST image, and use this information to revise the expected location of ASASSN-15og in our HST image.¹⁸ The revised absolute astrometry error radius is $\sqrt{{\sigma }_{\alpha }^{2}+{\sigma }_{\delta }^{2}}$, where σ is the standard deviation in the barycenter of ASASSN-15og as measured by Source Extractor in the Swift image (i.e., the square root of the variance or the second-order moment, X2_WORLD). The new expected location of ASASSN-15og is shown in Figure 1. Although the Swift image did not decrease the radius of the astrometric error circle, it is now centered on object A and rules out an association with object B.

ASASSN-15og was also observed by Gaia (as Gaia16ajy) and reported as a Gaia Alert at J2000 coordinates α = 03^h21^m0749, δ = −31°18′4594¹⁹ on 2015 October 29, about 2.5 months after the initial discovery (Brimacombe et al. 2015a). We obtained the Gaia coordinates for the nearby star from the Gaia Data Release 1 source catalog²⁰ and used the Gaia coordinates for the star and ASASSN-15og to revise the expected SN location in our HST image, as shown in Figure 1. The absolute astrometric uncertainty of the Gaia DR1 objects is extremely low (<0001), and the astrometric uncertainty on Gaia Alerts has been quoted to be similarly low (∼005),²¹ but we checked the absolute astrometric offset between these two catalogs ourselves. We searched the Gaia Alerts archive for objects in the vicinity of ASASSN-15og (using the maximum search radius of 20°), and found seven that appeared to be point-like nontransient sources, such as variable stars and quasars, that were also listed in the Gaia DR1 source catalog. We find that there is always a small offset in the coordinates for the same object in the Alerts and DR1 catalogs, and that this offset appears to be random and not systematic (i.e., it is not always in the same direction). We use the standard deviation of these offsets to calculate that the 1σ error radius in the expected position of Gaia16ajy is σ_Gaia = 014. In Figure 1, the radius of the Gaia circle is 3σ_Gaia, and we can see that object A is consistent with the expected location of ASASSN-15og.

We now endeavor to confirm that object A is an astrophysical point source and not, for example, the chance alignment of two CRs. With SN 2015cp, we obtained follow-up observations that confirmed the NUV source as a true detection, and an analysis such as presented below for ASASSN-15og was not necessary. As discussed in the final paragraph of this section, because ASASSN-15og had already exhibited signatures of CSM interaction, it was not prioritized for follow-up observations.

First, we checked the data-quality frames and determined that object A is not flagged as a possible CR. Then, we injected a population of simulated point sources, again using the TinyTim PSF for WFC3 as in Section 3.2, with total fluxes similar to that of object A into the two flc images. We used the mask frame to ensure that we only placed these simulated sources in clean pixels, and then we ran Source Extractor with the same settings as in Table 4. In Figure 2, we plot the measured flux radius versus the measured source flux for all objects detected in either flc image, using circles to represent injected PSFs, five-point stars for object A, and squares for all the rest, which are most likely to be CRs. Figure 2 demonstrates that object A has a flux radius similar to a TinyTim PSF and unlike a CR. It also shows that the flux of object A is very similar in both flc images and lies in a flux range that is relatively unoccupied by noninjected sources in the image. We made similar comparisons with other Source Extractor output parameters such as the FWHM intensity, aperture flux, and peak surface brightness (not shown), with similar results: it appears that object A is a real source for which we cannot rule out an association with the location of ASASSN-15og. Unfortunately, there is no previous HST imaging at these coordinates, and so we cannot say whether or not it is a pre-existing source unrelated to ASASSN-15og.

Zoom In Zoom Out Reset image size

To obtain accurate photometry for object A, we use the injected point sources to derive a correction between the true and measured flux for this image, similar to (but technically not) an aperture correction. For this we use Source Extractor’s FLUX_AUTO because it demonstrates the tightest correlation with input flux (better than the aperture fluxes we tested, for 3, 5, and 10 pixel radii). We use the image zero point from the header keywords and find that the observed apparent magnitude of object A is m_F275W = 23.94 ± 0.08 mag. The Galactic extinction for the coordinates of ASASSN-15og is A_F275W = 0.061 ± 0.002 mag (Schlafly & Finkbeiner 2011), and so the extinction-corrected apparent magnitude is m_F275W =23.88 ± 0.08 mag.

We also observed the location of ASASSN-15og with the Low Resolution Imaging Spectrometer at Keck Observatory on 2017 January 3, just under two months after the HST NUV image was obtained. Our B and R filter images do not show any point source at the site of the SN, but the location has a higher amount of host-galaxy surface brightness in the optical filters than in the NUV. We estimate that a point source of g ≈ 20 mag could have been detected at a signal-to-noise ratio of ∼1, given its location in the host galaxy, but this does not put any further constraints on the nature or brightness evolution of our HST detection.

This late-time detection of ASASSN-15og is not surprising: its peak brightness was overluminous (Gaia Alerts; Guillochon et al. 2017), and its classification optical spectrum exhibited a blue continuum and Balmer emission—all indicators of CSM interaction. ASASSN-15og was initially classified as an SN IIn/SN Ia–CSM ∼20 days after peak brightness (Monroe et al. 2015), and a reanalysis of its early-time data by Holoien et al. (2017) reclassified this event as an SN Ia–CSM (removing the SN IIn designation). We included a few such objects with early-onset CSM interaction in our target list as proof of concept because they have a much higher probability of yielding a late-time NUV detection. Since the main science goal of this project was to find and study SNe Ia that appeared normal at the time of discovery and up until regular monitoring ended, but then exhibited interaction only at late times, we did not trigger any further follow-up observations of ASASSN-15og.

3.3.2. SN 2015cp

In this section, we demonstrate our in-depth image review that confirmed the nondetections of ASASSN-15de, SN 2013dn, and SN 2014ch. The point-source limiting magnitudes for these images were obtained with the method discussed in Section 3.2.

3.4.1. ASASSN-15de

3.4.2. SN 2013dn

3.4.3. SN 2014ch

The NUV emission from the host galaxies of SN IIn 2014ab and 91T-like SNe Ia 2015bd and PS15sv is brighter than anticipated, as shown in Figure 3. For both SNe 2014ab and 2015bd, no images from the time of the SN are publicly available, so we cannot further revise the expected location or decrease the absolute astrometric error radius. SN Ia PS15sv was imaged by the Las Cumbres Observatory, and we use five of the best-seeing images from 2015 April in order to refine the expected location of the SN in our HST image.

Zoom In Zoom Out Reset image size

To detect sources in these images, we use mostly the same Source Extractor parameters as in Table 4, except for more aggressive deblending parameters and a smaller background mesh size to compensate for the relatively higher NUV background. All of the detected sources that are within the astrometric error circles for these three SNe have been confirmed to be unassociated with cosmic rays. However, they all also have an extended profile (FWHM ≳3 pixels), indicating that they are not point sources, but associated with the underlying host-galaxy emission, and therefore not suitable candidate detections of late-time interaction. We conclude that our HST NUV imaging for these three SNe has yielded nondetections.

To estimate the limiting magnitudes of these three images, we use the same methods of injecting and recovering fake implanted PSFs as described in Section 3.2, but with tweaked Source Extractor parameters. For these images, we only consider an injected source “recovered” if it is detected as a point source (i.e., with 1.0 ≤ FWHM ≤ 2.5), which was not a necessary condition for the other images’ limiting magnitude derivations (Section 3.2). The 50% detection limit for our SN 2014ab and SN 2015bd images is m_F275W = 25.5 mag, and for PS15dv it is m_F275W = 25.8 mag (reported in Table 3).

SN 2015cp is located 960 E and 996 N of its presumed host galaxy, a nearly face-on spiral with SDSS photometry g = 17.5 and r = 16.7 mag. To refine the host-galaxy redshift, on 2017 October 25 we obtained a low-resolution optical spectrum of the host with the Kast Spectrograph on the 3 m Shane telescope at Lick Observatory.

The Hα emission line is at an observed wavelength of λ = 6834 ± 2 Å, indicating a host-galaxy redshift of z = 0.0413 ± 0.0003, similar to that derived from the SN classification spectrum, z ≈ 0.038 (Frohmaier et al. 2016). The luminosity distance to the host galaxy of SN 2015cp, assuming a flat universe with H₀ = 70 km s⁻¹ Mpc⁻¹ and Ω_M = 0.3, is 167.5 Mpc. We adopt the values z = 0.0413 and d = 167.5 Mpc for SN 2015cp throughout this work.

The detection of SN 2015cp was first announced on 2015 December 28 by the PanSTARRS survey with a discovery magnitude of 18.8 (Frohmaier et al. 2016), and a second PanSTARRS epoch of i = 20.1 mag was obtained on 2016 January 23 (Huber et al. 2015; Guillochon et al. 2017). The designation for the SN in this survey is PS15dpq.

We found that SN 2015cp was also covered by the iPTF survey (iPTF15fel) with a partial light curve in g and R that extends from 2015 mid-December to 2016 mid-January; no other existing photometry was found for SN 2015cp. We fit the iPTF light curve using SiFTO (Conley et al. 2008) as shown in Figure 4, holding the redshift fixed at z = 0.041. Since all of the photometry is after the light curve’s maximum brightness, the time and magnitude at peak are not well constrained. The best fit for the date of peak brightness is 2015 November 18 ± 4 days; the fit template has a rise of 17 days (rest frame), and the estimated date of first light in the observer frame is 2015 October 28 ± 4.7 days. The best fit for the peak apparent magnitudes are m_g ≈ 17.5 ± 0.3 and m_R ≈ 17.1 ± 0.2 in the AB system, in the observer frame, after correcting for Milky Way extinction. This light-curve fit indicates an intrinsic peak brightness of M_B ≈ −18.5 ± 0.3 mag, a color at peak brightness of B − V ≈ 0.38 ± 0.07 mag, and a light-curve stretch factor of s = 1.13 ± 0.09 (where stretch values >1 indicate broader/brighter light curves; Perlmutter et al. 1997). If we include the correction for the known correlation between color and peak brightness (i.e., the βc correction, which accounts in part for dust extinction along the line of sight), we find M_B ≈ −19.6 mag, which is overluminous compared to normal SNe Ia (M_B ≈ −19 mag). The relatively large color correction implies that the light from SN 2015cp may have experienced significant dust extinction, which is somewhat unexpected given that SN 2015cp is well offset from its host galaxy, and may suggest extinction caused by CSM. However, because the photometry does not cover the light-curve peak, we cannot securely derive a value for dust extinction in the local environment and do not attempt to correct for it in later analyses (Sections 4 and 5).

Zoom In Zoom Out Reset image size

A spectrum of SN 2015cp was obtained on 2016 January 2 by PESSTO (Smartt et al. 2015) at a phase of 45 days after light-curve peak. The initial analysis of this spectrum classified SN 2015cp as a 91T-like SN Ia at z ≈ 0.038 with a spectroscopic phase of ∼40 days after peak brightness (Frohmaier et al. 2016); these are in agreement with both our light-curve fit and host-galaxy redshift. We used the online SN spectral classification tool GELATO²² (Harutyunyan et al. 2008) to verify this classification and found that it is 100% identified as an SN Ia. The top three best-fit spectral templates are all to SN 1998es, a 91T-like SN Ia, 46–58 days past light-curve peak (Ayani & Yamaoka 1998; Jha et al. 1998); no CSM interaction was reported for SN 1998es. The fourth best is to SN Ia–CSM PTF11kx 39 days past maximum light, which was near the time that CSM interaction started for PTF11kx (Dilday et al. 2012). The spectrum of SN 2015cp and examples of the best-fit GELATO templates are shown in Figure 4. We include for comparison a spectrum of SN Ia 2011by (Graham et al. 2015a), a twin of the prototypically normal SN Ia 2011fe, which exhibits narrower features that are stronger in the blue. Although the line widths and depths of SN 2015cp do appear to be more similar to those of SN Ia 1991T-like PTF11kx, we do not think that a normal SN Ia spectral match can be ruled out for SN 2015cp based on this PESSTO spectrum. We furthermore note that the overluminous peak brightness and late-type host galaxy are also consistent with the SN 1991T-like subtype (e.g., Sullivan et al. 2006; Howell et al. 2009).

This spectrum does not exhibit any Na i D λ5890 absorption, which, especially if it were blueshifted, would have indicated the potential presence of CSM. However, in this case, the absence of a visible Na i D absorption feature is not evidence that there is no CSM, because the optical spectrum is quite noisy. Furthermore, no obvious Na i D absorption features were seen in a spectrum of PTF11kx at similar phase and signal-to-noise ratio, and PTF11kx exhibited CSM interaction shortly thereafter (Figure 4; Dilday et al. 2012). Given the flux noise in the 45 day spectrum of SN 2015cp, we can rule out the presence of narrow Hα and Ca ii emission lines with the same luminosity as those exhibited by PTF11kx at a similar phase (i.e., as shown in Figure 4) with 10σ and 2σ confidence, respectively. This strongly suggests that for SN 2015cp, the CSM interaction had not yet begun 45 days after peak brightness.

After our HST NUV detection of SN 2015cp, we pursued ground-based optical spectroscopic observations to look for Hα emission, which would confirm that the NUV signal was caused by the SN ejecta interacting with CSM.

We first obtained 8 × 300 s r-band images on 2017 October 12 with the Palomar Observatory 60 inch telescope (the seeing was ∼186), at a phase of 694 days since light-curve peak brightness (30 days after the HST observation). SN 2015cp was not detected in a deep stack of these images, which has an upper limit of m_r = 23.6 mag. That same night, we obtained R-band imaging and a 2 × 600 s optical spectrum with the XShooter instrument (Vernet et al. 2011) on the Very Large Telescope (VLT) at Cerro Paranal in Chile. We detected SN 2015cp with low signal-to-noise ratio in the R-band image with an apparent magnitude of m_R ≳ 24 mag and detected emission lines from a point-like trace in the spectra.

The spectrum shown in Figure 5 exhibits both Hα and Ca ii emission. These lines are not associated with the nebular SN emission, which is dominated by forbidden lines from the nucleosynthetic products (Fe, Co, and Ni) and would be too faint to be detected by these observations. These lines of Hα and Ca ii emission revealed by the VLT spectrum provided the crucial confirmation that the observed NUV luminosity was associated with SN ejecta interacting with CSM that contains H. In order to monitor the evolution in the Hα emission, we obtained further VLT+XShooter spectra on 2017 November 16 and 21, shown in Figure 5.

Zoom In Zoom Out Reset image size

We also obtained a Keck spectrum of SN 2015cp on 2017 November 17 with the Low Resolution Imaging Spectrometer (LRIS; Oke et al. 1995). We acquired 3 × 1200 s exposures with a long slit of width 10, the 560 nm dichroic, the 400/8500 grating, and the 600/4000 grism. The data were processed using standard techniques optimized for Keck+LRIS by the CarPy package in Pyraf (Kelson et al. 2000; Kelson 2003). We performed flux calibration using sensitivity functions derived from each of the two standard-star observations made that night, which led to two reduced versions with different total Hα fluxes (Figure 5). Both standards were used simply as a conservative precaution, but this has revealed a systematic error in our spectral calibrations which we cannot explain—for example, the weather on Maunakea was stable and clear that night.²³ Since neither the absolute Hα flux measurements nor the shape of its flux decline curve is used in any of our analysis, this issue with the calibration of the Keck data from 2017 November 17 does not affect our results. On 2018 January 11, we obtained another 3 × 1200 s spectrum of SN 2015cp in a similar setup with Keck+LRIS, as shown in the bottom panel of Figure 5, which provided conclusive evidence that the Hα flux was declining.

The dominant feature in the optical spectra of SN 2015cp is the broad Hα emission line with an asymmetric profile that exhibits a blueshifted peak. This profile is consistent with the signature of dust formation in an emitting volume that is expanding: the receding material that generates the red half of the line experiences greater extinction because of the additional dust along the line of sight. We make direct measures of the peak velocity, FWHM, and integrated flux of the Hα emission line for all of our spectra. The latter is calculated by cutting out and interpolating over the pixels associated with host-galaxy emission (the very narrow features at v ≈ 0 km s⁻¹ in Figure 5) and performing a numerical integration on the calibrated one-dimensional (1D) spectrum. Since there is no continuum flux in our spectra, we do not need to perform a continuum subtraction prior to the integration. The resulting line parameters are listed in Table 5, and the flux evolution is plotted in Figure 6, where we can see that the Hα flux is declining by a factor of ∼3 in ∼90 days (L_Hα ∝ t^−8.6). Even considering the large (systematic and random) error bars on the flux measurements and the decline rate, this indicates an extremely rapid decline. For comparison, an ideal gas expanding as V ∝ t³ (V is volume and t is time) with an adiabatic equation of state P ∝ ρ^5/3 (P is pressure and ρ is mass density) would have its optically thin free–free luminosity decreasing as ${L}_{\mathrm{ff}}\propto {\rho }^{2}{T}^{-1/2}V\propto {t}^{-2}$.

Zoom In Zoom Out Reset image size

We do not detect Hβ λ4861 emission, but we can estimate an upper limit on the line intensity as three times the noise in that region. Using the first VLT spectrum obtained at +694 days, this leads to I(Hβ) < 0.3 × 10⁻¹⁷ erg s⁻¹ cm⁻² Å⁻¹. Assuming that there is no underlying continuum emission (or that it is the same in the regions of Hα and Hβ), this puts a lower limit on the Balmer decrement of D = I(Hα)/I(Hβ) > 3, which is consistent with Case B recombination in an ionized nebula and the large Balmer decrements seen in promptly interacting SNe Ia–CSM (Silverman et al. 2013b). However, because this is based on a noisy nondetection of Hβ, we do not rely on this estimate for any further analysis.

At the red end of the optical spectrum, we see the Ca ii near-infrared triplet, appearing as a blend of Gaussian-like features. Compared to the Hα emission line, we find that the Ca ii is a bit broader (FWHM ∼ 3000 km s⁻¹) and/or perhaps exhibits a slightly extended blue-side wing (at λ ≲ 8440, but it is unclear given the noise in flux) and clearly has an overall blueshift of ∼−1400 km s⁻¹ (significantly larger than Hα; Table 5). One explanation for a slightly extended blue-side wing might be the presence of O i λ8446, which we show as a dashed vertical line in Figure 5. However, for O i to contribute to the Ca triplet in this way, O i would have to be at rest velocity, and furthermore, O i is not typically seen in the late-time spectra of SN Ia–CSM like it is for SN II, as it is a common element in the ejecta of core-collapse SNe (Silverman et al. 2013b; Fox et al. 2015). We do not see the O i λ7774 Å line (Figure 5), and unfortunately, the VLT XShooter spectrum is too noisy in the region of O i λ11290 Å to confirm the presence of oxygen in the spectrum.

The additional blueshift of Ca ii compared to Hα is certain, but is a phenomena we cannot yet explain. Since Ca ii is at redder wavelengths than Hα, the extra blueshift is unlikely to be due to more severe effects of dust extinction. Furthermore, explaining the additional Ca ii velocity offset with dust extinction would require significant extinction on the near side of the gas (to produce the observed zero intensity at zero velocity). A potential explanation is that perhaps the Ca ii line is forming in a higher velocity region than Hα, but there is no physical motivation to expect this. An alternative interpretation is offered by Fransson et al. (2014), who presented a well-sampled, multiwavelength temporal series of observations for SN IIn 2010jl, which also exhibits blueshifted spectral features. Instead of dust extinction, Fransson et al. (2014) found it more likely that the emitting CSM has been accelerated by the SN’s radiation. We ultimately conclude that the SN 2015cp spectral data are insufficient to distinguish among these competing explanations. As a final note in support of dust extinction as the root cause of the observed blueshift for Hα, we note that other SNe Ia–CSM such as PTF11kx have exhibited increasing infrared emission after the ejecta pass through the CSM, indicating dust formation in the post-shock nebula (e.g., Graham et al. 2017).

The decline rate of the Ca ii near-infrared triplet’s flux is not as securely measured as that of Hα's flux owing to the noise at the red end of the spectrum, which prevents a detailed analysis. Although the Ca ii appears to be declining more slowly than Hα, because of the large uncertainties on the flux measurements, we cannot rule out that they are co-evolving and that the two features are both connected to the shock evolution. This is discussed further by H18 (their Figure 2 provides a direct comparison of the Ca ii and Hα flux decline rates).

These spectra also potentially exhibit a Mg i λ5174 feature with a broad or blueshifted profile, but this has a very low signal-to-noise ratio that prohibits further analysis.

After confirming the interaction origin of our NUV detection, we submitted an HST Director’s Discretionary Time (DDT) proposal requesting one NUV spectrum with the STIS MAMA G230L grating and a time series of NUV photometric monitoring with the WFC3 F275W filter. We requested the spectrum to determine the nature of the NUV emission, and we suspected that the NUV emission would be line-dominated for two reasons: (1) our initial analysis of the NUV detection indicated that bremsstrahlung (free–free) and synchrotron emission could not account for the observed luminosity (see also 5.3), and (2) the Mg ii λ2800 resonance line was covered by HST's F275W filter for this SN. We also proposed for several closely spaced F275W imaging epochs followed by a slower cadence in order to determine the decline rate of the emission, because this would give some indication of the timescale of the CSM interaction and the radial extent of the material. Our request for an STIS spectrum was granted, with the photometric time series conditional on the confirmed presence of Mg ii (GO-15407; PI Graham).

On 2017 November 25, at a phase of 738 days since the light-curve peak, HST executed our program and obtained a 12,991 s spectrum with the STIS NUV MAMA G230L grating and the 52 × 05 slit. Our choice of the 05 width was to ensure that SN 2015cp was within the slit at the time of the exposure; as described in the STIS Data Handbook, “Once the telescope acquires its guide stars, your target will be within 02–03 of the aperture center” (Bostroem & Proffitt 2011). Unfortunately, absolutely no source is seen at the expected location in the processed two-dimensional and 1D spectra. In Figure 7, we show the extracted 1D spectrum (from the STIS pipeline’s x1d task), and it is clearly dominated by noise with no emission. To estimate our detection limit for a Gaussian-shaped emission line of Mg ii λ2800, we inject simulated Gaussians into the original STIS data. We use a range of FWHM and peak velocities that are similar to our observations of Hα, and a range of fluxes that would yield an apparent magnitude similar to and fainter than our initial NUV detection of m_F275W = 24.3 mag. We then smooth the spectrum with a Savitsky–Golay filter of 5 Å (i.e., less than our expected FWHM) and attempt to fit a Gaussian function using randomized initial parameters and scipy.optimize.curve_fit. We consider the planted emission line successfully recovered if the fit has a peak wavelength λ_peak < σ/2 and also has a similar FWHM and peak intensity; this process is illustrated by the planted-and-recovered Gaussians in Figure 7.

Zoom In Zoom Out Reset image size

From this procedure, we report an upper limit on the emission-line brightness at +738 days of m_F275W ≈ 25.1 mag, the magnitude at which our emission-line recovery fraction drops to 50% (and we checked that this result is the same if we do not smooth the spectrum). As reported in Section 3.3.2, our NUV detection at +664 days had m_F275W ≈ 24.3 mag. If 100% of that flux was from an emission line of Mg ii λ2800 declining at the same rate as Hα and Ca ii, then the expected integrated flux at +738 days is m_F275W ≈ 25.3 mag. This is slightly fainter than our upper limit from the STIS spectrum, and so our nondetection remains consistent with our hypothesis that Mg ii line emission is the underlying cause of our NUV detection.

If the NUV luminosity at +664 days was the result of continuum emission, then we would not have detected it in our STIS spectrum at +738 days unless the brightness had increased to m_F275W ≤ 22.3 mag (i.e., the magnitude that would provide an average signal-to-noise ratio of ∼1 per resolution element in the spectrum). Additionally, we note that if the NUV luminosity at +664 days was due to high-temperature blackbody emission with f_NUV ≈ 3 × 10⁻¹⁸ erg s cm² Å, then its optical component would be undetectable in our optical spectra (Figure 5). Therefore, we conclude that our observations cannot rule out continuum emission as the underlying cause of our NUV detection, and furthermore note that it is implausible given that the optical spectrum is line-dominated.

Unfortunately, the photometric monitoring aspect of our HST DD proposal was conditional on detecting the Mg ii line, so we were unable to proceed with our program to obtain multiple epochs of photometry to further understand the NUV luminosity evolution.

In this section, we use our NUV and optical follow-up observations to deduce the most likely characteristics of SN 2015cp’s Hα and NUV emission and their evolution, and to make a preliminary physical interpretation of the amount and location of CSM in its progenitor system. Detailed physical modeling of the CSM properties of SN 2015cp can be found in H18, who incorporate radio and X-ray follow-up observations of SN 2015cp to produce quantitative constraints.

As described in Section 4.2, the PESSTO spectrum of SN 2015cp is best fit by templates from the 91T-like subclass of SNe Ia, such as SN Ia 1998es and SN Ia–CSM PTF11kx. The SN ejecta of PTF11kx were observed to interact with CSM starting ∼60 days after its explosion, marked by the appearance of H and Ca emission features (Dilday et al. 2012). The evolution of PTF11kx’s spectral features was monitored for hundreds of days afterward (Silverman et al. 2013a), generating a significantly better temporal series of data for PTF11kx than we have for SN 2015cp. For this reason, we base most of our preliminary physical interpretation of SN 2015cp’s CSM on a comparison with PTF11kx.

In Section 4.2, we presented the optical light curve of SN 2015cp, from which we determined the date of explosion to be 2015 October 28 ± 4.7 days. The fact that the 2016 January 2 spectrum does not exhibit any signature of CSM interaction (e.g., a broad FWHM ≈ 2000 km s⁻¹ Hα emission line, as seen for PTF11kx at a similar phase in Figure 4) indicates that the ejecta of SN 2015cp did not impact the CSM earlier than ∼60 days after explosion. Assuming a maximum ejecta speed of v_ej,max ≈ 30,000 km s⁻¹, this indicates that the CSM was at a distance of at least R_CSM > 10¹⁶ cm from the WD in the SN 2015cp progenitor system. Since the ejecta of PTF11kx were observed to begin interacting with CSM ∼60 days after explosion, we conclude that the CSM in the system of SN 2015cp was more distant than that of PTF11kx, though perhaps only marginally.

Silverman et al. (2013a) showed that the luminosity of the broad Hα line of PTF11kx increased for ≳60 days after the onset of CSM interaction, and then exhibited a plateau of 200–400 days before declining. The start of the Hα luminosity decline has been interpreted as indicating the end of interaction (Silverman et al. 2013a; Graham et al. 2017), in other words, the time at which the ejecta had fully passed through the CSM. However, the spectral time series of PTF11kx is too sparse to accurately measure the Hα luminosity decline rate or the true duration of the plateau. We have shown in Figure 6 that SN 2015cp has a declining L_Hα throughout our observations, which indicates that our discovery occurred well after the ejecta–CSM interaction occurred, and perhaps very soon after the shock had swept through the CSM, given the rapidity of the decline (as discussed in Section 4.3). The Hα light curve of PTF11kx may also have declined more rapidly in the ∼100 days after its plateau than at much later times (e.g., Figure 4 of Graham et al. 2017), but the Hα light curve is too sparsely sampled to be sure. A direct comparison of the L_Hα light curves of SN 2015cp and PTF11kx is presented by H18 (their Figure 2).

We find that the Hα line profiles of SN 2015cp and PTF11kx are similar in width and shape between the two events, but that the integrated luminosity is different—for SN 2015cp, L_Hα is about an order of magnitude lower than that of PTF11kx at a similar phase. Whether this is due to a less dense CSM for SN 2015cp or a difference in the CSM radial extents for the two events is not possible to distinguish from our data. Unfortunately, no NUV observations of PTF11kx were obtained, so we cannot extend our comparison of these two SNe Ia–CSM any further.

As a final aspect of our physical interpretation for SN 2015cp, we consider the possible mechanisms that could have produced the observed NUV emission and what they may reveal about the nature of the CSM in the SN 2015cp system. The main open question is whether the NUV signal was from a Mg ii λ2800 emission line or from a continuum process like optically thin bremsstrahlung. As described in Section 4.4, our STIS nondetection does not allow us to rule out either. However, one argument in favor of line emission is that an improbably large CSM mass, ∼10 M_⊙, would be required to produce enough continuum emission (see Section 5.3). Another argument in favor is that the models of Chevalier & Fransson (1994) show that the Mg ii λ2800 line can be luminous enough to account for the NUV emission—specifically, up to eight times more luminous than Hα when the ejecta of an SN II interact with a slow, dense wind. This is not an identical scenario to SNe Ia, but similar physical mechanisms are at play. These models are consistent with our observation that ${L}_{\mathrm{Mg}{\rm{II}}}/{L}_{{\rm{H}}\alpha }\lt 6$ (based on our late-time HST imaging and ground-based optical spectra for SN 2015cp, presented above). Unfortunately, the requisite modeling to evaluate the density and extent of CSM associated with such NUV line emission remains to be done and is beyond the scope of this work.

We have demonstrated that the ensemble NUV through optical observations of SN 2015cp are consistent with an overluminous, possibly SN 1991T-like SN Ia, for which the ejecta interacted with CSM containing H no earlier than ∼60 days after explosion. This sets the inner CSM radius of SN 2015cp to be R_CSM > 10¹⁶ cm, farther than that of SN Ia–CSM PTF11kx (but perhaps not by much). Our HST NUV detection at 686 days after explosion sets the maximum inner radius of the CSM to be R_CSM < 10¹⁷ cm. During the ∼120 days after our NUV detection, which itself was nearly two years after explosion, our optical and NUV follow-up observations reveal rapidly declining Hα and Ca ii emission, which is typically associated with the interaction of CSM and SN ejecta. The rapid decline suggests that we caught SN 2015cp soon after that emission peaked in brightness, which in turn implies that the interaction-driven peak in Hα luminosity was probably fainter than that of PTF11kx and that the CSM of SN 2015cp has either a lower density or a smaller radial extent than in PTF11kx. Finally, we argue that our NUV photometric detection was likely produced by line emission rather than thermal continuum emission (although the latter could not be ruled out by our observations), but that without more detailed modeling of line-emission mechanisms, our NUV detection alone cannot place any independent constraints on the CSM properties in the SN 2015cp progenitor system.

In order to analyze our full sample of late-time NUV observations for this diverse set of SNe, we must convert our NUV apparent magnitudes or magnitude limits to intrinsic spectral luminosities and the dates of our HST observations to phase in days after SN peak brightness. For each of our 70 undetected targets, we convert the limiting magnitudes that represent the 50% detection probability, as listed in Table 3, into an upper limit on the intrinsic NUV luminosity by correcting for Milky Way extinction (Schlafly & Finkbeiner 2011) and using the distances to each object from Table 1. For our two detected SNe, ASASSN-15og and SN 2015cp, we convert the apparent NUV magnitudes into an intrinsic NUV luminosity in the same way. For all objects, we use the HST observation dates from Table 3 and the dates of peak brightness from Table 6, which come from a combination of private (e.g., PTF) and public (e.g., The Astronomer’s Telegrams) resources (column 3 of Table 6), to estimate the SN’s phase at the time of every HST observation. The phases and derived spectral luminosity limits are also listed in Table 6. In Figure 8, we plot our NUV luminosity limits versus phase as inverted gray triangles.

Zoom In Zoom Out Reset image size

5.1.1. Host-galaxy Extinction Corrections

5.1.2. ASASSN-15og and SN 2015cp

5.1.3. Comparing NUV Emission for SNe

We use our HST images’ detection limits (Section 3.2), and our detection of SN 2015cp, to constrain the fraction of SN Ia systems that have CSM (f_CSM), the occurrence rate of CSM interaction, and observational attributes of the NUV emission’s peak, duration, and decline rate. Since our targeted sample of SNe Ia is overrepresentative of the subtypes that are more likely to be associated with CSM interaction (Section 2.1), our results will serve as an upper limit. In the following paragraphs, we introduce the key components to this analysis—the probabilistic expectation value, control time, detection efficiency, and NUV light-curve parameters. For simplicity, in this section we assume a single shell of CSM.

Expected Number of Detections—The number of late-time NUV detections expected in our survey, N_exp, is represented by

$$\begin{eqnarray}&&{N}_{\exp }={f}_{\mathrm{CSM}}\displaystyle \sum _{i=1}^{i={N}_{\mathrm{SNIa}}}P(\mathrm{occ}\leqslant {t}_{\mathrm{obs}})\times P(\det | \mathrm{occ}),\end{eqnarray} \tag{ 1 }$$

where f_CSM is the primary unknown we constrain with this analysis: the fraction of SNe Ia in our sample that have CSM within R_CSM ≈ 3 × 10¹⁷ cm. Moore & Bildsten (2012) showed that ∼10¹⁷ cm is about the maximum distance of a CSM shell that is formed by recurrent nova systems (see their Figure 4). By design, 3 × 10¹⁷ cm is the distance traveled by SN ejecta with v ≈ 20,000 km s⁻¹ in +1700 days, which is the maximum phase probed by our survey.

The sum term in Equation (1) calculates the total number of detections that we would expect if all of our targeted SNe had CSM within R_CSM ≈ 3 × 10¹⁷ cm, and it has two components. The first, P(occ ≤ t_obs), is the probability that the interaction occurred before the date of our observation, t_obs, given that there was CSM within R_CSM ≈ 3 × 10¹⁷ cm. We assume that the probability density distribution for CSM distance is flat, although this depends on the properties of the recurrent nova system and thus the distribution of (for example) mass-transfer rates in those systems (Moore & Bildsten 2012). The second, $P(\det | \mathrm{occ})$, is the probability that the NUV emission would be detected in our HST image, given that it occurred,

$$\begin{eqnarray}&&P(\det | \mathrm{occ})={\rm{\Delta }}t/{t}_{\mathrm{obs}},\end{eqnarray} \tag{ 2 }$$

where Δt is the control time as described below. The value of $P(\det | \mathrm{occ})$ is always ≤1.

We exclude from Equation (1) the five SNe with classifications that include SN IIn, which means they exhibited clear and strong signatures of CSM interaction <100 days after optical peak, and the three classified at early times as SNe Ia–CSM (Table 1), because our goal is to constrain the fraction of SNe Ia in our sample with late-onset interaction.

Control Time—The concept of control time, Δt, also referred to as the visibility window, is the length of time over which a variable or transient object is detectable, in days:

$$\begin{eqnarray}&&{\rm{\Delta }}t={\int }_{0}^{\infty }\eta (m(t)){dt},\end{eqnarray} \tag{ 3 }$$

where η(m) is the survey’s detection efficiency (defined below) and m(t) is a fiducial light curve of the transient or variable object (defined below). For use in Equation (2), we impose that the maximum value for the control time is the SN’s phase at the time of our observation: Δt ≤ t_obs.

Detection Efficiency—The detection efficiency of an imaging survey, η(m), is the probability of detecting a source of apparent magnitude m with a given threshold of confidence. In Section 3.2, we determined the detection efficiency of our survey by injecting point sources with a variety of fluxes into our HST images and calculating the fraction that were detected at the 2σ level (Table 4). The functional form of η(m) has two components: a bright-end flat regime in which η(m) = η_max for m < ${m}_{{\eta }_{\max }}$ and a faint-end decline in which the probability of detection drops with an approximately linear slope down to η = 0 at m = m₀. This functional form holds for all of our HST images, and the average parameter values are η_max ≈ 0.954, ${m}_{{\eta }_{\max }}$ ≈ 25.6 mag, a faint-end slope of −0.15 per 0.1 mag, and m₀ ≈ 26.3 mag.

The NUV Light Curve—The full evolution of NUV emission from late-onset CSM interaction for SNe Ia might be heterogeneous, as it is for SNe IIn, but this diversity has not yet been observed. Instead, we must estimate a probable functional form for an NUV light curve; to do this, we base our fiducial NUV light curve of CSM interaction, m(t) in Equation (3), on the analytical models of Harris et al. (2016). They show that the radio light curve from CSM interaction exhibits a sharp rise when the ejecta impact the CSM, flattens out while the ejecta pass through the CSM, and then rapidly declines after shock breakout. Based on this, we use a simple functional form for the spectral luminosity L_NUV:

$$\begin{eqnarray}{L}_{\mathrm{NUV}}(t)=\left\{\begin{array}{ll}{L}_{\max } & \mathrm{if}\ t\leqslant { \mathcal W }\\ {L}_{\max }\times {{\rm{\Phi }}}^{(t-{ \mathcal W })/100} & \mathrm{if}\ t\gt { \mathcal W }.\end{array}\right.\end{eqnarray} \tag{ 4 }$$

The time t is in days after the ejecta impact the CSM, and L_NUV is in units of erg s⁻¹ Hz⁻¹. In this model, the luminosity exhibits a plateau of width ${ \mathcal W }$ days followed by a fractional flux decline of Φ every 100 days. This parameterization is not a direct derivation from the radio light curves, but adopted as a simple way to represent the two main light-curve features: the width of the plateau (${ \mathcal W }$) and the rate of the decline (Φ). To use this light-curve model in Equation (3), we convert the spectral luminosity to an apparent magnitude in the HST F275W filter.

Results—To demonstrate the methodology, we calculate N_exp for a hypothetical scenario in which all of our targeted SNe Ia had CSM within R_CSM ≈ 3 × 10¹⁷ cm (f_CSM = 1), and all exhibited emission like that of SN 2015cp: a peak intrinsic spectral luminosity of L_NUV = 7.6 × 10²⁵ erg s⁻¹ Hz⁻¹ that lasts for ${ \mathcal W }=200$ days, followed by a fractional flux decline of Φ = 0.3 every ∼100 days. These values come from our analysis of SN 2015cp in Section 4: the peak-luminosity estimate is a lower limit based on our single detection, the width estimate is based on our comparison of SN 2015cp to PTF11kx, and we assume that the NUV and Hα fluxes would have a similar decline rate.

The control time for this fiducial, SN 2015cp-like light curve in our HST survey is Δt ≈ 330 days, and if all of our targets exhibited emission like SN 2015cp, we would have expected N_exp = 17.4 detections in our survey. The fact that we only detected one suggests that the fraction of our targets with CSM is closer to f_CSM = 0.06. Of our assumptions for the three light-curve parameters—the peak, width, and decline rate—the width is the one that is most uncertain: it is a time span based on a single epoch and essentially unconstrained. To illustrate the effect of our width assumption on our results, ${ \mathcal W }=50$ days (a minimum based on the light-travel time across a shell of R ∼ 10¹⁷ cm) results in f_CSM = 0.09, and ${ \mathcal W }=600$ days (a maximum considering the nondetections <50 days after peak brightness, and t_obs = 664 days) results in f_CSM = 0.03.

In reality, the amount and distribution of CSM in SN progenitor systems will have some intrinsic diversity, and Harris et al. (2016) showed that, for example, thicker CSM shells result in a wider light curve and denser shells in a brighter maximum luminosity. We therefore extend our analysis to incorporate a suite of light curves with peak luminosities in the range of $24.0\leqslant \mathrm{log}{L}_{\mathrm{NUV}}\leqslant 27.5$ erg s⁻¹ Hz⁻¹, widths in the range of $20\lesssim { \mathcal W }\leqslant 1000$ days, and fractional decline rates in the range of 0.1 ≤ Φ ≤ 0.8. These ranges represent the extent of what might be both physically possible and detectable by our survey; in particular, the shallower decline rates represent the slower NUV light-curve evolution seen for SNe IIn (e.g., Figure 8).

In Figure 9, we show the number of expected detections in this light-curve parameters space, normalized by the fiducial results of f_CSM = 0.06. In the top panel, we hold the fractional flux decline rate fixed at Φ = 0.3 and show N_exp in the width–peak plane, and in the bottom panel we hold the light-curve width fixed at ${ \mathcal W }=200$ days and show N_exp in the decline-peak plane. Red boxes mark the location of the fiducial SN 2015cp-like light-curve in this parameter space. We find that even when the brightest, longest, slowest light curves of CSM interaction are considered, we would expect N_exp < 3 detections with our HST survey (for f_CSM = 0.06). Gehrels (1986) showed that the Poisson limits at 2σ confidence for a single detection are 0.173–3.3, which indicates that this entire range of light-curve parameters is statistically consistent with our observations.

Zoom In Zoom Out Reset image size

Our single-epoch survey is unable to put any stronger constraints on the fraction of distant CSM in SN Ia progenitor systems, or the occurrence rates or light-curve parameters of its associated NUV emission. As a final caveat, we note that because our SNe Ia were targeted for being more likely to be associated with CSM interaction (Section 2.1), even the constraints that we can provide should be considered an upper limit.

We can use the models of Harris et al. (2016, hereafter H16) to convert our upper limits on the NUV luminosity into upper limits on the CSM mass. As shown by H16, the interaction shock heats up the CSM and the ejecta, increasing the volume of hot gas until the ejecta have swept through all of the CSM, at which point the shock-accelerated CSM rapidly expands and cools. These models assume that the CSM has low density (ρ < 10⁻¹⁷ g cm⁻³) to avoid significant photon trapping. A low density is also appropriate for our systems: a dense CSM at a distance of R_CSM > 10¹⁶ cm would imply a larger mass than a red giant companion star could be expected to generate. Additionally, such high densities and large masses of CSM would cause significant dust extinction and narrow, blueshifted absorption features in the spectra obtained around the time of peak light, which are not generally seen for SNe Ia.

In order to derive upper limits on the CSM mass, we assume that the time of our observations, t_obs, corresponds to the time of the peak of the CSM interaction-driven NUV light curve, t_p, and use the H16 models to constrain the bremsstrahlung luminosity from the CSM at the moment the shock reaches the outer edge of the CSM. This assumption does not lead to the most conservative estimates on the amount of CSM, but was chosen as an intuitive and practical option for interpreting our single-epoch nondetections. If we instead assume that t_p was any time in the hundreds of days before t_obs, the upper limits on the peak luminosity (and CSM mass) remain essentially unbounded; to avoid this contrived assumption, future surveys for late-onset CSM interaction need to include multiple epochs of imaging. Owing to the assumption of t_p = t_obs, we omit PTF11kx from this analysis because we know it was well past the peak of its interaction at the time of our observation (t_obs = 2134 days), and we also omit any targeted SN with a classification of Type IIn because this spectral type indicates that it exhibited clear signatures of strong CSM interaction around the time of peak brightness (when it was classified). Any emission from the reverse shock would also contribute to our observed luminosity—although it would be a small contribution, including it would lower the CSM mass required to generate a given luminosity.

First, we will summarize some of the main results of H16 that are required for this analysis. We start by setting up the physical parameters for the CSM and ejecta density, location, mass, and volume before moving on to modeling the NUV emission from the shocked CSM. H16 described a CSM shell of constant density ρ_CSM with an inner radius of R_in, an outer radius of R_out, and a fractional extent of

$$\begin{eqnarray}&&{F}_{R}\equiv {R}_{\mathrm{out}}/{R}_{\mathrm{in}}.\end{eqnarray} \tag{ 5 }$$

We consider shells with 1.1 ≤ F_R ≤ 10, where the lower value is like a nova shell (Moore & Bildsten 2012), and the upper value represents a very thick shell. For PTF11kx, observations indicated F_R ≈ 5 (e.g., Graham et al. 2017). A strong shock forms in the CSM when ejecta of approximately ${\rho }_{\mathrm{csm}}$ reaches the inner edge of the CSM. As in H16, this sets the impact time ${t}_{\mathrm{imp}}$, because the outer ejecta density profile is described by

$$\begin{eqnarray}&&{\rho }_{\mathrm{ej}}(r,t)=\displaystyle \frac{0.124\,{M}_{\mathrm{ej}}}{{\left({v}_{t}t\right)}^{3}}{\left(\displaystyle \frac{r}{{v}_{t}t}\right)}^{-10},\end{eqnarray} \tag{ 6 }$$

where r is the radial distance from the SN center, t is time since explosion, M_ej is the total mass of the ejecta, and v_t is the transition velocity between the outer ejecta and inner ejecta. For a typical SN Ia, the transition velocity is v_t = 1.104 × 10⁹ cm s⁻¹, and the mass M_ej = 1.38 M_⊙ = 2.74 × 10³³ g. For the family of models presented by H16, which have ${\rho }_{\mathrm{ej}}({R}_{\mathrm{in}},{t}_{\mathrm{imp}})={\rho }_{\mathrm{csm}}/0.33$, this profile leads to the relation

$$\begin{eqnarray}&&{R}_{\mathrm{in}}=3.43\times {10}^{9}\,{t}_{\mathrm{imp}}^{7/10}\,{\rho }_{\mathrm{csm}}^{-1/10}.\end{eqnarray} \tag{ 7 }$$

H16 found that the time at which the shock overtakes the edge of the CSM shell (t_p) is related to the time of impact and the fractional extent by

$$\begin{eqnarray}&&{t}_{p}=0.983\,{t}_{\mathrm{imp}}\,{F}_{R}^{9/7}.\end{eqnarray} \tag{ 8 }$$

Because these models are appropriate for low-density shells, the evolution assumes a gamma-law equation of state with adiabatic index γ_ad = 5/3. Thus, the strong shock increases the CSM (and ejecta) density by a factor of 4: shocked CSM has ${\rho }_{s}=4{\rho }_{\mathrm{csm}}$, and the initial volume of the shell is simply expressed by

$$\begin{eqnarray}&&{V}_{\mathrm{csm}}=(4\pi /3){R}_{\mathrm{in}}^{3}\,({F}_{R}^{3}-1).\end{eqnarray} \tag{ 9 }$$

At the time we are considering, t_p, the CSM volume is ${V}_{s,p}={V}_{\mathrm{csm}}/4$ owing to shock compression.

Now we turn to H16's models for the bremsstrahlung emission from the shocked CSM at t_p, the peak time of the interaction-driven luminosity. For the densities and shell extents that we are considering, the bremsstrahlung process is very securely in the optically thin regime—and recall that we are calculating at the time that all of the CSM has been shocked. In the optically thin regime, the spectral luminosity (erg s⁻¹ Hz⁻¹) can be estimated from the emissivity (ε_ν) and emitting volume V_s to be ${L}_{\nu }={\varepsilon }_{\nu }{V}_{s}$.

The emissivity in cgs units (erg s⁻¹ cm⁻³ Hz⁻¹), from Equation 5.14(b) of Rybicki & Lightman (1979), is

$$\begin{eqnarray}&&{{ \mathcal E }}_{\nu }^{{ff}}\approx 6.8\times {10}^{-38}\,{Z}^{2}\,{n}_{e}\,{n}_{i}\,{T}^{-1/2}\,{e}^{-h\nu /{kT}}\,{\bar{g}}_{{ff}},\end{eqnarray} \tag{ 10 }$$

where Z is the charge on the ion, n_e and n_i are the particle densities of electrons and ions (respectively), T is the temperature, and the frequency is ν ≈ 1.11 × 10¹⁶ s⁻¹ (from the pivot wavelength of the HST WFC3 F275W filter, λ = 2704 Å). We set the thermally averaged Gaunt factor ${\bar{g}}_{{ff}}=1$, appropriate for an order-of-magnitude estimate at near-optical wavelengths. The temperature of the shocked CSM depends on its density, but it will be high (10⁸–10⁹ K) owing to its low density and the high shock speed. Given that the energy of an NUV photon is hν ≈ 6.6 × 10⁻¹¹ erg, the exponential term in the emissivity is approximately unity. We furthermore assume the simple case of complete ionization of H-dominated material in the emitting region, such that Z = 1 and n_e = n_i = ρ_s/m_p = 4 ${\rho }_{\mathrm{csm}}$/m_p, where m_p is the proton mass.

Using ${\rho }_{s}=4{\rho }_{\mathrm{csm}}$, ${M}_{\mathrm{csm}}={\rho }_{\mathrm{csm}}{V}_{\mathrm{csm}}$, and Equations (5)–(9), we find that the NUV bremsstrahlung spectral luminosity of the fully shocked CSM shell at the moment the shock reaches its outer edge is

$$\begin{eqnarray}&&{L}_{\nu }\approx 1.63\times {10}^{-31}\,{T}_{s}^{-1/2}\,{t}_{\mathrm{obs}}^{-3}\,{M}_{\mathrm{csm}}^{17/7}\,y({F}_{R}),\end{eqnarray} \tag{ 11 }$$

where $y{({F}_{R})={F}_{R}^{-3/7}(1-{F}_{R}^{-3})}^{-10/7}$. The $y({F}_{R})$ term is dominated by the first factor for ${F}_{R}^{3}\approx 1$ and by the second factor for ${F}_{R}^{3}\gg 1$. As in Equation (10), all units are in cgs. From this equation, it is clear that a maximum allowed CSM mass can be found from a luminosity upper limit, for a known t_obs, once values for T_s and F_R are assumed. For example, an observation of an NUV luminosity limit ${L}_{\nu ,\mathrm{NUV}}={10}^{25}\,\mathrm{erg}\,{{\rm{s}}}^{-1}\,{\mathrm{Hz}}^{-1}$ rules out the presence of a nova-like shell (F_R = 1.1) more massive than 5 M_⊙ (assuming that the post-shock temperature is T_s = 10⁸ K).

The behavior of Equation (11) is shown in the left panel of Figure 10 for two of the SNe in our sample: the positive detection of SN 2015cp and our earliest nondetection, iPTF16abc—which was well studied for its peculiarity at early times (Ferretti et al. 2017; Dhawan et al. 2018; Miller et al. 2018). The tighter mass constraint for iPTF16abc is mostly because the luminosity is limited to be fainter than the detection of SN 2015cp (see Figure 8), but also because the observation time of iPTF16abc is earlier and thus requires a closer, denser shell that would be more luminous for a given mass than a more distant shell. The two curves in the left panel of Figure 10 represent the extrema of the full range of constraints provided by all of our NUV HST observations.

Zoom In Zoom Out Reset image size

The final results of this analysis—the upper limits on the CSM mass for all of our observed SNe Ia, based on our models of the bremsstrahlung (free–free) luminosity from the hot shocked CSM—are shown in the right panel of Figure 10. We illustrate the full range of allowed F_R for a given SN, after imposing ${t}_{\mathrm{imp}}$ ≥ 60 days since explosion, which puts a lower limit on R_in (to incorporate the fact that none of these SNe Ia exhibited the signatures of interaction during their initial observations). In the right-hand plot of Figure 10, circles mark the mass limit for an F_R = 1.1 shell, and the vertical lines extend up to the mass limit at the maximum possible F_R for each SN (i.e., the horizontal axis of the left-hand plot). We note that the CSM densities that these limits imply are consistent with the optically thin assumption (limiting how high the CSM density can be) and the assumption that SNe Ia surely do not have ejecta faster than 45,000 km s⁻¹ (limiting how low the CSM density can be).

Figure 10 demonstrates that, when we only consider bremsstrahlung emission from the shocked CSM, our survey was sensitive to CSM shells with masses as low as a few M_⊙ for our closest targets. Based on this analysis alone, it would appear that our survey was insensitive to small amounts of CSM: several tenths of a solar mass in some cases, but >1 M_⊙ in most cases, could have remained undetected. However, as we explain below, we find that this insensitivity is more related to our assumption of bremsstrahlung emission for the physical modeling than to any fault of this survey’s observational depth or design.

For example, in Figure 10, we see that the lower limit on the free–free CSM mass of SN 2015cp is ∼7 M_⊙, which is an unreasonably high CSM mass. For comparison, the CSM mass limit for SN 2015cp based on radio follow-up observations is M_csm ≲ 0.5 M_⊙ (H18; but we note that obtaining that limit depended heavily on the NUV discovery and optical follow-up data). This discrepancy illustrates why we suspect the NUV emission to originate from atomic line emission (Mg ii at 2800 Å) rather than continuum emission. Unfortunately, line emission is much more complicated than bremsstrahlung to approach theoretically—it requires many more assumptions about the state of the gas. Such a tool for translating our upper limits into CSM masses based on line emission is not yet in hand, and we consider its development beyond the scope of this work.

One possible alternative approach assumes that the ${\rm{H}}\alpha$ line is formed through radiative recombination during the cooling of shocked gas (Ofek et al. 2013, for example), but as this results in CSM masses (or magnesium abundances) that are too high to be physically feasible, we do not attempt to apply it to our inferred NUV lines in this work. An appropriate and accurate model of the line formation for the detached CSM shell scenario, like that presented by Chevalier & Fransson (1994) for SNe II interaction with a wind, will (in the future) allow us to utilize the results of our survey to deduce more sophisticated and stringent mass limits. We cannot apply the models of Chevalier & Fransson (1994) directly to our observations because the hydrodynamics of SNe Ia are very different from those of SNe II that they have modeled: SNe Ia have 1/10th the ejecta mass of SNe II, the mass and distribution of their CSM are quite different, and SNe Ia have steeper density gradients in their outer layers and their fastest moving ejecta travel at about twice the speed compared to those of SNe II. Although Chevalier & Fransson (1994) found a general trend that SNe with a steeper gradient for the ejecta’s radial density distribution lead to more luminous emission from the reverse shock (i.e., their Figure 3), this cannot be directly extended to our discussion about a potential contribution to the NUV emission from the reverse shock in the ejecta of SN 2015cp. Additionally, timescales for the decline of NUV emission dominated by contributions from the reverse shock are expected to be much longer, perhaps years compared to the few-week decline we infer from our HST follow-up nondetection for SN 2015cp. In lieu of such models for SNe Ia, we again stress the advantage of multiwavelength observations that include X-ray and radio bands (as done by H18) for constraining the CSM properties.