We present the first observations of a Type I superluminous supernova (SLSN) at $gtrsim 1000$ days after maximum light. We observed SN 2015bn using the Hubble Space Telescope Advanced Camera for Surveys in the F475W, F625W and F775W filters at 721 days and 1068 days. SN 2015bn is clearly detected and resolved from its compact host, allowing reliable photometry. A galaxy template constructed from these data further enables us to isolate the SLSN flux in deep ground-based imaging. We measure a light curve decline rate at $>700$ days of $0.19 pm 0.03$ mag (100 d)$^{-1}$, much shallower than the earlier evolution, and slower than previous SLSNe (at any phase) or the decay rate of $^{56}$Co. Neither additional radioactive isotopes nor a light echo can consistently account for the slow decline. A spectrum at 1083 days shows the same [O I] and [Ca II] lines as seen at $sim300-400$ days, with no new features to indicate strong circumstellar interaction. Radio limits with the Very Large Array rule out an extended wind for mass-loss rates $10^{-2.7} lesssim dot{M}/v_{10} lesssim 10^{-1.1}$ M$_odot$ yr$^{-1}$ (where $v_{10}$ is the wind velocity in units of 10 km s$^{-1}$). The optical light curve is consistent with $L propto t^{-4}$, which we show is expected for magnetar spin-down with inefficient trapping; furthermore, the evolution matches predictions from earlier magnetar model fits. The opacity to magnetar radiation is constrained at $sim 0.01$ cm$^2$ g$^{-1}$, consistent with photon-matter pair-production over a broad $sim$GeV-TeV range. This suggests the magnetar spectral energy distribution, and hence the missing energy leaking from the ejecta, may peak in this range.
Download