One Thousand Days of SN2015bn: HST Imaging Shows a Light Curve Flattening Consistent with Magnetar Predictions

We imaged SN 2015bn using the HST Advanced Camera for Surveys Wide Field Channel⁶ on 2017-06-01.4 and 2018-06-22.3 (all dates in UT), corresponding to 721 and 1068 days after maximum light in the rest-frame of SN 2015bn. Visits consisted of one orbit per filter in F475W, F625W, and F775W, corresponding closely to g, r, and i bands, where we expect most of the strong emission lines (Nicholl et al. 2016b, 2018; Jerkstrand et al. 2017). Each image contained four dithers in a standard box pattern.

We retrieved the fully processed and drizzled images from the Mikulski Archive for Space Telescopes. Figure 1 shows the combined three-color images. SN 2015bn is clearly visible as a point source superimposed on its host galaxy. We removed the host using a galaxy model constructed with galfit (Peng et al. 2002), fitting a Sérsic profile while masking the pixels that were clearly dominated by SN 2015bn. There were no significant differences between the fits obtained in the individual epochs. Subtracting the model from the HST images resulted in a clean SN detection with minimal galaxy residuals, as shown in Figure 1. We then performed point-spread function (PSF) photometry with daophot, and applied standard zeropoints.⁷ We verified that the zeropoints were consistent between the two epochs (to within <0.02 mag) using 16 stars from the Pan-STARRS Data Release 1 catalog (Flewelling et al. 2016).

Zoom In Zoom Out Reset image size

We obtained ground-based imaging on 2017-02-01.3 and 2018-03-18.7 using the Low Dispersion Survey Spectrograph 3 (LDSS3) on the 6.5-m Magellan Clay telescope. Each observation consisted of 10 × 300 s dithered r-band exposures, which we reduced in pyraf. From the ground, SN 2015bn appears entirely blended with its (much brighter) host. Subtracting the galfit model derived from the HST data, after convolving to the ground-based resolution using hotpants,⁸ we isolated the SN light and performed PSF photometry, determining the zeropoints using the Pan-STARRS catalog.

Our photometry is plotted in Figure 2, along with earlier g, r, and i data from Nicholl et al. (2016a, 2016b). The latest points are fainter than the peak by a factor ≈1500, but a flattening in the light curve beyond ∼500 days is immediately apparent. The new data have been submitted to the Open Supernova Catalog (Guillochon et al. 2017).

Zoom In Zoom Out Reset image size

We observed SN 2015bn spectroscopically on 2018-07-08.9 (1083 rest-frame days after maximum) using LDSS3. The data were reduced in pyraf, with flux calibration achieved using a standard star. The spectrum is shown in Figure 3. The mean airmass during the observation was 1.6, and the spectrum redward of ∼7500 Å is contaminated by noise residuals from sky subtraction.

Zoom In Zoom Out Reset image size

Although the spectrum is dominated by the host, the strongest emission lines from SN 2015bn appear to be visible above the galaxy light. We subtract a model for the host continuum (Nicholl et al. 2016a) and compare to the most recent prior spectrum (at 392 days after maximum; Nicholl et al. 2016b), scaled to match the latest HST observations. We find that the broad feature at 6300 Å is consistent with predictions for [O i] λ6300, while a tentative feature at 7300 Å matches [Ca ii] λ7300. This indicates that the lines have changed little, despite a gap of 691 days. Our new spectrum likely represents the oldest spectroscopic detection with respect to explosion for any SLSN: the normalized phase is t/t_d = 13.5 in the terminology of Nicholl et al. (2018), where t_d = 80 days is the decline timescale of the light curve.

We also obtained spectra of three galaxies that apparently neighbor SN 2015bn (labeled 3–5 in Figure 1). We find that they are a background group at z = 0.353 unrelated to SN 2015bn. The bright (M_r ≈ −21) spiral galaxy (2) has a spectrum from the Sloan Digital Sky Survey Data Release 7 (Abazajian et al. 2009) that indicates z = 0.1118, similar to SN 2015bn.

The relative line-of-sight velocity between this galaxy and the SN host is cΔz = 540 km s⁻¹, while their projected separation is ≈56 kpc. These values are similar to the Magellanic Clouds relative to the Milky Way, and the absolute magnitude (M_r = −16.4), physical size, star formation rate, and metallicity of the host (Nicholl et al. 2016a) are all similar to the SMC. Thus, the host and its bright neighbor appear to be a close analog of the MW-SMC system.

Chen et al. (2017) found that the host of one SLSN, LSQ14mo, was in a likely interacting system with a projected separation of 15 kpc, and proposed that this could increase the likelihood of SLSNe by triggering vigorous star formation. The brightest and bluest pixels in our HST images, likely corresponding to the highest star formation rate, actually appear to be on the other side of the galaxy, though we cannot exclude comparable star formation at the position of SN 2015bn until the SLSN has completely faded.

We observed SN 2015bn using the Karl G. Jansky Very Large Array (VLA) in B configuration, on 2017 November 10 (867 rest-frame days after maximum).⁹ SN 2015bn was not detected to 3σ limiting flux densities of 48 μJy in K band (21.8 GHz) and 63 μJy in Ka band (33.5 GHz).

Light echoes occur when light emitted earlier in the SN evolution is reflected into our line of sight by nearby dust sheets, giving an apparent luminosity boost after a light travel time. For nearby SNe, this is readily identifiable through a change in the spatial emission profile, but at the distance of SN 2015bn, 1 lt-yr corresponds to only ∼10⁻⁴ arcsec. An echo beginning ∼2 years after explosion could roughly match the late-time brightness if it was ≈8 mag fainter than the light curve peak. Lunnan et al. (2018) recently detected the first light echo in a H-poor SLSN, iPTF16eh, via a Mg ii resonance line.

There are several issues with interpreting the behavior of SN 2015bn as an echo. First, the luminosity of an echo is expected to evolve as t⁻¹ (e.g Graur et al. 2018), which is flatter than what we observe. Second, the spectrum is consistent with a typical SLSN nebular spectrum, whereas an echo should contain features from earlier phases, when the SN was brighter. However, we caution that the spectrum is noisy and dominated by host galaxy light.

Finally, dust is more efficient in reflecting blue light, which changes the observed colors. We measure g − i = 0.45 ± 0.24 at 1068 days, which is consistent with the color at 300 days (g − i = 0.35 ± 0.17) but not with the peak (g − i = −0.27 ± 0.02). A similar finding applies to g − r and r − i. We therefore conclude that an echo cannot explain the slow evolution.

Follow-up of nearby SNe at ≳900 days has revealed evidence for the decay chain ⁵⁷Ni → ⁵⁷Co → ⁵⁷Fe, in both core-collapse (Seitenzahl et al. 2014) and Type Ia SNe (Shappee et al. 2017; Graur et al. 2018). While the relative abundance of ⁵⁷Ni is typically low (⁵⁷Ni/⁵⁶Ni ≲ 0.05), the long lifetime of ⁵⁷Co (half-life ≈ 272 days) means that it eventually comes to dominate over ⁵⁶Co.

The decay slope for ⁵⁷Co is 0.28 mag(100 d)⁻¹, which is comparable to our light curve, but still somewhat faster. A small contribution from the slower reaction ⁵⁵Fe → ⁵⁵Mn (half-life ≈ 1000 days) could help to mitigate this. Seitenzahl et al. (2014) also looked for signatures of ⁶⁰Co and ⁴⁴Ti in SN 1987A, but the half-lives of these species are too long (5–60 years) to be relevant to SN 2015bn yet.

The more significant problem for this scenario is that the pseudobolometric luminosity of SN 2015bn at 900 days is 10^40.8 erg s⁻¹, i.e. 400–4000 times greater than SNe Ia at the same phase (Graur et al. 2018). The required ⁵⁷Co mass is ≳6 ${\text{}}{M}_{\odot }$. We are not aware of any explosion model capable of producing this; even the most massive pair-instability models from Heger & Woosley (2002) synthesize an order of magnitude less ⁵⁷Co (while making 40 ${\text{}}{M}_{\odot }$ of ⁵⁶Co). For a solar ratio of ⁵⁷Co/⁵⁶Co = 0.023 (Lodders 2003), the implied ⁵⁶Co mass would be >260 ${\text{}}{M}_{\odot }$.

Assuming a velocity of ∼7000 km s⁻¹ (Nicholl et al. 2016a, 2016b), we see that the ejecta expand to a radius ≈6 × 10¹⁶ cm within 1000 days, and the fastest ejecta likely reach ∼10¹⁷ cm. Yan et al. (2017) found that up to ∼15% of SLSNe encounter hydrogen-rich CSM at ∼10¹⁶ cm, as indicated by the sudden appearance of broad hydrogen emission lines in their spectra, while Lunnan et al. (2018) identified a circumstellar shell at ≳10¹⁷ cm around iPTF16eh from its light echo. Thus, interaction with a massive CSM at a similar radius could be a plausible luminosity source for SN 2015bn.

However, none of the interacting events in Yan et al. (2017) showed a shallow light curve resembling SN 2015bn, though the interaction in those events occurred much earlier (100–250 days) when the SLSNe were ∼1–2 orders of magnitude brighter. Chen et al. (2018) recently studied SN 2017ens, another SLSN that developed strong and broad Hα emission at ≳100 days, finding its light curve was essentially flat at this phase.

We examine the Hα region of our spectrum in Figure 3. We subtract a model consisting of the scaled 392 days spectrum and a linear host continuum, and fit the Hα and [N ii] lines with Gaussian profiles. A satisfactory fit is obtained with the width fixed at the instrumental resolution; i.e., the lines are unresolved, and no broad component is present above the level of the noise. The flux in Hα is 1.6 × 10⁻¹⁶ erg s⁻¹ cm⁻², consistent with host emission (Nicholl et al. 2016a).

While Hα is generally the strongest line in SNe interacting with hydrogen-rich material, interaction with hydrogen-free material is more difficult to exclude. Ben-Ami et al. (2014) detected narrow [O i] λ5577 emission from SN 2010mb, and proposed that it was a signature of interaction. We do not observe this line in SN 2015bn to a limit of ≲4 × 10³⁷erg s⁻¹, which is ∼10–100 times fainter than the line in SN 2010mb up to one year after explosion. A possible caveat is that this line is only predicted to be strong at densities >10⁷ g cm⁻³.

Late-onset interaction in other events has been interpreted as a collision with a detached shell, but a slow decline could also result from an extended dense wind. Figure 4 shows predicted radio emission for SNe interacting with winds of different densities (Kamble et al. 2016), which we compare to our VLA limit and an earlier limit from Nicholl et al. (2016a). Parameterizing the wind mass-loss rate as $\dot{M}/{v}_{10}$, where v₁₀ is the wind velocity in units of 10 km s⁻¹, the combined limits at 1–3 years rule out winds with ${10}^{-2.7}\lesssim \dot{M}/{v}_{10}\lesssim {10}^{-1.1}$ ${\text{}}{M}_{\odot }$ yr⁻¹. For a typical Wolf–Rayet wind velocity ∼1000 km s⁻¹, this corresponds to ${10}^{-4.7}\,\lesssim \dot{M}\lesssim {10}^{-3.1}$ ${\text{}}{M}_{\odot }$ yr⁻¹, excluding a wind significantly more dense than those from SN Ic progenitors (e.g., Berger et al. 2002; Soderberg et al. 2006; Drout et al. 2016).

Zoom In Zoom Out Reset image size

Comparing to models for the optical luminosity from Nicholl et al. (2016a), we can rule out most of the parameter space where the light curve peak can be powered by interaction with a dense wind, and also disfavors this as the primary late-time power source.

The most popular model for SLSNe is a magnetar engine. At late times the engine power decays as L ∝ t^−α; a standard magnetic dipole has α = 2. A long-standing prediction is that SLSNe should eventually track this power law. While many SLSN light curves have been observed to flatten with time, late observations have generally been either similar to ⁵⁶Co decay (Inserra et al. 2013) or of insufficient signal-to-noise ratio (Lunnan et al. 2016) to make strong statements.

In Figure 2, we plot representative curves for α = 2 and α = 4. The best-fitting power law at 200–1100 days has α ≈ 3.8, steeper than a standard dipole. However, it is expected that the energy available from spin-down is not completely thermalized at late times; assuming that this energy is injected primarily as high-energy photons, the optical depth in the expanding ejecta decreases with time as τ ∝ t⁻² (Wang et al. 2015; Chen et al. 2015). Including this "leakage" term gives

$$\begin{eqnarray}&&L\propto {t}^{-2}(1-{e}^{-{{kt}}^{-2}})\approx {{kt}}^{-4},\end{eqnarray} \tag{ 1 }$$

where $k=3{\kappa }_{\gamma }{M}_{\mathrm{ej}}/(4\pi {v}_{\mathrm{ej}}^{2})$ is the trapping coefficient, M_ej and v_ej are the mass and velocity of the ejecta and κ_γ is the opacity to high-energy photons. The second equality comes from a Taylor expansion applicable at late times. Thus a realistic power law is not α = 2, but rather α ≈ 4, close to what we observe.

We model the full light curve of SN 2015bn using mosfit (Guillochon et al. 2018). SN 2015bn has previously been fit using this code and a magnetar model by Nicholl et al. (2017), who described the methodology. The result is shown in Figure 5. Given that new data comprise only ≈1% of the total light curve points, it is unsurprising that the fit is unchanged with respect to Nicholl et al. (2017). We find a spin period P/ms = 2.32 ± 0.22, magnetic field $\mathrm{log}(B/{10}^{14}{\rm{G}})=-0.51\pm 0.09$, and ejecta mass $\mathrm{log}({M}_{\mathrm{ej}}/{M}_{\odot })=1.04\pm 0.03$. More interesting is that the best fit to the first 400 days gave a reasonably accurate prediction of the evolution at >1000 days. The model matches the data at 721 days, and agrees to better than a factor two at 1083 days, though the later data appear systematically above the fit. The previous modeling suggested κ_γ ∼ 0.01 cm² g⁻¹, which we confirm here.

Zoom In Zoom Out Reset image size

3.4.1. Is There a "Missing Energy Problem"?

The mechanisms discussed in Sections 3.2–3.4 assume that energy deposition is instantaneous. However, if the heat source is coupled to the ejecta through ionization and recombination, this assumption holds only if the recombination timescale is shorter than the heating timescale, which may not be true at late times when the ejecta density is low. This process of "freeze-out" can result in a light curve tracking the recombination rate instead of the heating rate (Fransson & Kozma 1993; Fransson & Jerkstrand 2015).

Following Kerzendorf et al. (2017) and Graur et al. (2018), we parameterize freeze-out as a luminosity source that evolves as t⁻³ (i.e., in proportion to the density, assuming constant expansion). Graur et al. (2018) defined t_freeze,50 as the time when freeze-out accounts for half of the emission. If freeze-out dominates by ∼700 days, we find t_freeze,50 ∼ 400 days. This is much earlier than in SN 1987A and a number of nearby SNe Ia, for which the timescales are typically ≳800 days (Fransson & Kozma 1993; Graur et al. 2018). It therefore seems unlikely that freeze-out alone can account for the flattening, but more detailed modeling is required here.