THE ULTRAVIOLET-BRIGHT, SLOWLY DECLINING TRANSIENT PS1-11af AS A PARTIAL TIDAL DISRUPTION EVENT

The PS1 telescope has a 1.8 m diameter primary mirror that images a field with a diameter of 33 (Hodapp et al. 2004) onto a total of 60 4800 × 4800 pixel detectors, with a pixel scale of 0258 (Tonry & Onaka 2009). A more complete description of the PS1 system, hardware and software, is provided by Kaiser et al. (2010). The nightly PS1 Medium Deep Survey (MDS) observations are obtained through a set of five broadband filters, designated as g_P1, r_P1, i_P1, z_P1, and y_P1, with a typical cadence of 3 d between observations in g_P1r_P1i_P1z_P1. Although the filter system for PS1 has much in common with that used in previous surveys, such as the SDSS (Ahn et al. 2012), there are differences, with further information on the passband shapes described by Stubbs et al. (2010). Photometry is reported in the "natural" PS1 system, m = −2.5log (f_ν) + m', with a single zeropoint adjustment m' made in each band to conform to the AB magnitude scale (Tonry et al. 2012). PS1 magnitudes are interpreted as being at the top of the atmosphere, with 1.2 airmasses of atmospheric attenuation being included in the system response function.

PS1 data are processed through the Image Processing Pipeline (IPP; Magnier 2006) on a computer cluster at the Maui High Performance Computer Center. The pipeline runs the images through a succession of stages, including flat-fielding ("de-trending"), a flux-conserving warping to a sky-based image plane, masking and artifact removal, and object detection and photometry. Transient detection using IPP photometry is carried out at Queen's University Belfast. Independently, difference images are produced from the stacked nightly MDS images by the photpipe pipeline (Rest et al. 2005) running on the Odyssey computer cluster at Harvard University. The discovery and data presented here are from the photpipe analysis.

We first detected PS1-11af in g_P1r_P1 images obtained on the night of 2010 December 30 after non-detections in our first observations of that field for the observing season on 2010 December 15/16 (g_P1r_P1i_P1). The transient rose to a peak in late January of 2011 and slowly faded thereafter, remaining detectable in PS1 imaging when observations ceased at the end of April due to solar conjunction. The mean position of PS1-11af was α = 9^h57^m26815, δ = +03°14'0100 (J2000), with an uncertainty of 01 in each coordinate. False-color images of PS1-11af near maximum light and its host galaxy are shown in Figure 1. The blue color of the transient relative to its red host galaxy is immediately apparent.

Zoom In Zoom Out Reset image size

We construct deep template observations from the pre-outburst images and subtract them from the PS1 observations using photpipe (Rest et al. 2005). Details of the photometry and generation of PS1 transient light curves are given by Rest et al. (2013) and Scolnic et al. (2013). The final PS1-11af photometry, after correction for E(B − V) = 0.027 mag of Galactic extinction (Schlafly & Finkbeiner 2011; Schlegel et al. 1998), is given in Table 1 and shown in Figure 2. Fourth-order polynomial fits to the g_P1 and r_P1 light curves near peak give dates for maximum light at Modified Julian Dates (MJDs) of 55581 ± 2 days (=2011 January 20), which we adopt as the time of peak throughout this paper.

Zoom In Zoom Out Reset image size

The field of PS1-11af was monitored in the near-ultraviolet band (NUV; Morrissey et al. 2007) as part of the GALEX Time Domain Survey (TDS; Gezari et al. 2013) from 2011 January 31 to 2011 February 16. The transient was detected in each observation during this period, while nothing was detected at the position of PS1-11af in either of the NUV or far-ultraviolet (FUV) bands in prior GALEX observations in 2008 and 2010. We perform aperture photometry of PS1-11af during outburst and list the results in Table 1. Further details of the GALEX survey design and data products were presented by Gezari et al. (2013).

In addition, we observed the host of PS1-11af on 2013 May 20 in J and H for a total of 1200 and 1254 s, respectively, using the FourStar Infrared Camera (Persson et al. 2008) on the 6.5 m Magellan Baade telescope. The images were flat fielded, sky subtracted, and stacked with standard tasks in IRAF¹⁸ using the instrument pipeline. The photometry was calibrated using 2MASS stars in the field.

We obtained four epochs of spectroscopy of PS1-11af in outburst between 2011 January 13 and 2011 June 10 using the Low Dispersion Survey Spectrograph-3 (LDSS3; an updated version of LDSS-2, Allington-Smith et al. 1994) on the 6.5 m Magellan Clay telescope, the Blue Channel Spectrograph (BC; Schmidt et al. 1989) on the 6.5 m MMT, the Inamori-Magellan Areal Camera and Spectrograph (IMACS; Dressler et al. 2006) on the Magellan Baade telescope, and the Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) on the 8 m Gemini-South telescope. A full log with details of the observations is given in Table 2. Conditions were clear for the initial LDSS3 and BC observations, but poor and variable seeing compromised the IMACS spectra. We combine the IMACS spectra on consecutive nights to increase the signal-to-noise ratio (S/N).

After PS1-11af had faded away, we returned to obtain spectroscopy of the host galaxy using similar BC and LDSS3 setups as those used for the original observations, including the same slit position angles (PAs) to within 5°. We also obtained spectroscopy with LDSS3 using a significantly redder setup, designed to cover Hα at the rest frame of the host galaxy. A few additional host spectra were obtained at PA = −10° for reasons discussed below but only when that angle was close to the parallactic angle.

We also obtained several images as part of the acquisition process using LDSS3, IMACS, and GMOS. We process the two-dimensional frames using standard tasks in IRAF. Multiple attempts to obtain the GMOS spectrum over the course of several days were aborted due to bad weather so we only obtained a g'r'i'z' acquisition sequence on each date. We stack the images in each filter and report the average dates in Table 1. We subtract the PS1 templates in the appropriate filters from the acquisition images and calibrate the photometry to PS1 stars in the field.

We reduce the spectroscopic data using standard tasks in IRAF along with our own IDL procedures to apply a flux calibration and correct for telluric absorption. The longslits were always aligned within 10° of the parallactic angle to mitigate the possible effects of differential atmospheric dispersion (Filippenko 1982). We did not use order-blocking filters, so second-order light contamination is a concern with a source this blue. Our early LDSS3 observation was taken using the standard 075 longslit located 2' blueward of the center of the field combined with the VPH-all grism. Past experience, along with tests using order-blocking filters on blue standard stars, has demonstrated that this instrument setup combination suffers very little contamination from second-order light. More generally, we use observations of both relatively blue (sdO spectral types) and red (F-type) standard stars taken both with and without order-blocking filters to define the flux calibrations for the BC and Magellan spectra. Despite our best efforts, we caution that some contamination may still be present at observed wavelengths longward of ~8000 Å. For GMOS-S, we use archival observations of EG21 to define the flux calibration. The final spectra are presented in Figure 3. All spectra exhibit absorption from the Ca ii H+K doublet at redshift z = 0.405, which we refine in Section 3.1 to z = 0.4046 and adopt as the host redshift throughout this paper.

Zoom In Zoom Out Reset image size

We observed the field of PS1-11af with the Karl G. Jansky Very Large Array (VLA; Dougherty & Perley 2010) three times. The first epoch was 2011 March 29.03 (Project 10A-214; PI: Soderberg), and subsequent epochs were obtained beginning 2012 January 7.38 (Project 11B-192; PI: Chomiuk) and 2013 May 31.96 (Project 13A-437; PI: Soderberg). Our observations are summarized in Table 3. The first observation was conducted with the old VLA system with two 128 MHz windows centered at 4.8 and 5.0 GHz. The latter two observations were conducted with the new WIDAR correlator (Perley et al. 2011) at a mean frequency of 5.9 GHz (lower sideband frequency centered at 5.0 GHz; upper sideband frequency centered at 6.75 GHz).

In all epochs, we use standard data reduction procedures in AIPS (Greisen 2003). We use 3C286 for bandpass and flux calibration, and J1024−0052 for gain calibration. We flag and excise channels affected by radio frequency interference, which results in an effective bandwidth of ~1.7 GHz. A ~35 mJy source 4' from the position of PS1-11af is used for self-calibration. We do not detect significant radio emission from PS1-11af in any epoch and set 3σ limits of 51, 30, and 45 μJy, respectively, from the root-mean-squared (rms) flux values of the images.

It is obvious from the strength of the 4000 Å break that the spectrum shown in Figure 4 is that of an older stellar population. The Balmer lines are only present in absorption, and there are no emission lines of any type, while absorption lines from several metals are strong. We cross-correlate this spectrum with the single stellar population age models of Bruzual & Charlot (2003). All models that are good fits have ages >1 Gyr, with the best fit at solar metallicity being 2.5 Gyr. The best-fit cross correlations give a precise redshift of z = 0.4046 ± 0.0001.

We use the FAST program of Kriek et al. (2009) to simultaneously fit the spectrum and photometry to a suite of solar metallicity stellar population models from Bruzual & Charlot (2003) with more complicated star formation histories. The best fit is overplotted in red and uses an exponentially declining star formation rate (SFR) with an e-folding timescale of 400 Myr and an age of 2.5 Gyr. Only the r_P1 and i_P1 points were used to scale the spectrum, so the good agreement in the other bands is reassuring. The total stellar mass is 7 × 10⁹ M_☉ when scaled just to the photometry in the inner aperture. The photometry in the outer aperture is ~0.6 mag brighter, so if all of the flux is assigned to the host, then the total stellar mass is closer to 1.2 × 10¹⁰ M_☉.

The difference spectrum shown at the bottom of Figure 4 shows no strong deviations between the model and the observed spectrum. In particular, no emission lines are visible. We use the difference spectrum and observed errors to set 3σ upper limits on the ([O ii] λ3727, Hα) emission fluxes of (5.2, 4.8) × 10⁻¹⁸ erg cm⁻² s⁻¹ in 10 Å bins around the rest wavelengths of the lines. According to the calibration of Kennicutt (1998), these correspond to upper limits on the SFR of the host galaxy of (0.04, 0.02) M_☉ yr⁻¹. Similarly, we also use the non-detections in the UV by GALEX and the calibrations of Kennicutt (1998) to set 5σ upper limits on the SFR of 0.8 M_☉ yr⁻¹. Combined, these indicate that there is no sign of star formation or any young stellar population in the host galaxy at the position of PS1-11af. In addition, we inspect our spectra of the host galaxy obtained at the PA of −10°. While the surface brightness of the extension is too low to obtain spectra with a good S/N in our limited exposure time, inspection of the two-dimensional frames shows no obvious emission lines.

An outburst in a persistently accreting AGN could masquerade as a true transient. However, the non-detection in the UV by GALEX argues strongly against the presence of a luminous unobscured AGN in the nucleus. The stacked non-detections in the GALEX bands correspond to absolute magnitudes M_UV > −17 mag or UV continuum luminosities νL_ν 6 × 10⁴² erg s⁻¹. Also, the difference spectrum at the bottom of Figure 4 exhibits no emission lines from the broad-line region (BLR) of an AGN. The spectra of both obscured and unobscured AGN could also manifest strong forbidden emission lines from a narrow-line region (NLR), but those are not present either. The 3σ limit on the [O iii] λ5007 luminosity is <1.4 × 10³⁹ erg s⁻¹, which falls below any of the optically selected type II AGNs at similar redshifts in the zCOSMOS survey (Bongiorno et al. 2010). However, we cannot exclude the possibility of a low-luminosity AGN, as samples in the local universe extend to significantly lower emission-line luminosities than our limits for the host of PS1-11af (Ho 2008; Hao et al. 2005). All of these upper limits would be made weaker by the presence of strong dust obscuration in the nucleus, but the blue color of the transient (Figure 3; see below) is incompatible with a large dust column along the line of sight.

The deep non-detection of the host galaxy in the UV prior to the detection of PS1-11af also represents a strong argument against normal accretion rate fluctuations in an AGN. As described by Gezari et al. (2012), quasars and AGN exhibit variability of a lower amplitude in GALEX observations than the >2.5 mag increase in brightness of PS1-11af relative to quiescence. In the full GALEX TDS observations (Gezari et al. 2013), most sources with such large amplitude NUV flares are classified as cataclysmic variables or M-dwarf flares. The only extragalactic sources having similarly large amplitude flares after non-detections in quiescence were the TDE PS1-10jh and a few SNe (Gezari et al. 2012).

Another possibility is that PS1-11af is an outburst from a blazar (Urry & Padovani 1995), but we would expect to detect a blazar in the radio. Near the peak of the outburst, PS1-11af had a V-band magnitude of ~21.5, corresponding to νL_ν ≈ 3 × 10⁴³ erg s⁻¹. The unified blazar SEDs of Fossati et al. (1998) predict radio fluxes in the range 0.4–15 mJy for different ranges of radio loudness. Even the faintest of these is ~8 times brighter than our 3σ upper limit in the first epoch. The broad UV absorption features of PS1-11af would also be unexpected in the synchrotron-dominated spectra of a blazar.

As a final check against AGN-like variability, we present difference light curves in five filters for PS1-11af from all four years of PS1 observations obtained to date in Figure 5. This light curve shows no sign of variability or any long-term trend during the other three observing seasons to date. The host galaxy was also observed by SDSS a decade prior to our detection of PS1-11af (object ID: SDSS J095726.82+031400.9), and the model magnitudes reported by Ahn et al. (2012) from 2001 February 20 are consistent within 1σ of our PS1 photometry in the large aperture (Table 4).

Zoom In Zoom Out Reset image size

The fast rise in g_P1 of PS1-11af is especially notable in this context. The host flux in the narrow aperture was g_P1 = 23.42 mag, but PS1-11af rose on a timescale of less than a month to a peak of g_P1 ≈ 21.5 mag, implying a rise of more than 2 mag (the quiescent flux is clearly dominated by star light; Figure 4). The characteristic amplitude of quasar photometric variability on these timescales is only ~0.1 mag, with the amplitude of variability increasing on longer timescales (e.g., Webb & Malkan 2000; MacLeod et al. 2012a). This is very different from the long-term PS1-11af light curve, which is consistent with a constant in the observing seasons lacking the transient.

The observed spectra of PS1-11af shown in Figure 3 exhibit Ca ii H+K absorption from the host galaxy as well as several undulations that correspond to similar features in the host spectrum (Figure 4). A comparison of the host galaxy photometry (Table 4) and the transient light curve (Table 1) reveals that even near maximum light we should expect a significant contribution from the host galaxy to the observed fluxes, particularly at the redder wavelengths. We now use the information about the host galaxy from Section 3.1 and the photometric properties of PS1-11af from Section 4 to subtract the host galaxy contribution from the observed spectra and isolate the spectrum of the transient.

The host galaxy is spatially resolved while the transient is not, so seeing variations mean that the amount of galaxy light relative to the transient in the spectroscopic slit aperture cannot be reliably determined directly from the photometry. Instead, we fit the observed spectra to model the host contribution. We initially model each epoch of spectroscopy as a linear sum of the best-fit host model from Section 3.1 and a BB with T_BB equal to the value determined above from the GALEX+PS1 SED fits and determine the best-fit scale factors for each component. We previously used a similar procedure to subtract the host galaxy of PS1-10jh from its spectra (Gezari et al. 2012). We then repeat our subtraction procedure using a power-law continuum with α = 0.73 and find statistically superior fits. At each epoch, the scaling factors for the amplitudes of the host galaxy model are identical for the BB and power law models to within 1%–3%, indicating that our subtraction procedure is not sensitive to this choice for the transient SED, although the scaling factors do change if T_BB or α are varied. An example fit to the first PS1-11af spectrum is shown in Figure 9. The depths of the absorption lines and amplitudes of the continuum undulations due to the host in the model spectrum (magenta line) match those in the data very well.

Zoom In Zoom Out Reset image size

Several features of the fitting procedure are worth noting. The first is that we do not include any constraint from the observed photometry (except implicitly through our choice of the power-law index or T_BB) due to the possibility that our absolute spectrophotometry is unreliable because of clouds or slit losses. However, the green squares overplotted on Figure 9 represent the g_P1r_P1i_P1z_P1 photometry interpolated to the date of the spectrum. The excellent agreement of the photometry with both the amplitude and color of the scaled power law is apparent. This gives us confidence that the spectrophotometry is correct and that our procedure to scale and subtract the host is working satisfactorily. This can be seen in another way in Figure 7, where we overplot the galaxy-subtracted spectra on Days −5 and +24 along with the photometry. Although we apply the indicated multiplicative offsets for clarity, no relative normalization factors were applied between the photometry and the spectroscopy at the same epoch. Again, the colors and normalization of the galaxy-subtracted spectra are in excellent agreement with the photometry.

Second, the spectra on the first two epochs exhibit a blue excess relative to the host plus BB model that can be seen at wavelengths below ~3200 Å in Figure 7. This is not simply a consequence of using an incorrect (too low) T_BB. If we allow T_BB to vary in our fits, we can find better fits to match the overall shape of the observed spectra if T_BB is in the range 25,000–30,000 K. However, the fits compensate for the bluer assumed BB color by increasing the amplitude of the host contribution to match the red flux. After subtraction of the new best-fit host contribution, the derived transient spectrum then has a color that is too blue relative to the colors measured from the photometry. Equivalently, the values for T_BB allowed by the optical photometry (Figure 8) are lower than those required to remove the blue/UV excess flux. The simplest explanation is that the blue excess is real and the SED of PS1-11af is not that of a pure single-temperature BB. Therefore, in the remainder of this paper, we use the statistically preferred power-law model to determine the proper scaling factors for the host galaxy. We do not claim that the true SED is a power law, just that a power law is a better approximation over the limited wavelength range of our spectra for the purpose of scaling and removing the host contribution. We also note that the shape of the excess is inconsistent with Balmer continuum emission, which would be expected to peak near 3650 Å.

Another consideration is that we subtract the spectral-synthesis model for the host galaxy spectrum but the real spectrum could be different. The excellent fit of the host model in Figure 4 demonstrates that any such error is small. We repeat the subtraction procedure using our actual host galaxy spectrum and find no significant difference in the subtracted spectra, including the presence of the UV excess in the fits with a BB continuum. However, the subtracted spectra are noticeably noisier, especially at shorter wavelengths, so we use the model-subtracted spectra in all subsequent analysis.

After determining the relative amplitude of the host galaxy contribution for each epoch,¹⁹ we subtract the appropriately scaled host galaxy model and present the resulting spectral sequence in Figure 10. The spectra of PS1-11af are very blue, with the Day −5 spectrum being well fit by a power law with α = 0.75 ± 0.02, consistent with the values measured from the photometry, as shown in the top panel of Figure 8. No clear emission features (either narrow or broad) are present in any of our spectra, unlike the broad He ii emission seen from PS1-10jh (Gezari et al. 2012) or the broad Hα emission detected from SDSS TDE2 (van Velzen et al. 2011a). The Day +24 spectrum exhibits two strong absorption features shortward of 3000 Å, with minima near 2450 and 2680 Å. The reddest of these is clearly not present on Day −5 in our only other spectrum with overlapping wavelength coverage. We note that these features are definitely real (they are apparent even in the two-dimensional spectral frames) and are largely unaffected by any details of the host galaxy subtraction procedure. The host galaxy has very little contribution at these wavelengths (e.g., Figure 9), and the features are present prior to host subtraction in Figure 3.

Zoom In Zoom Out Reset image size

PS1-10jh was a well-observed TDE discovered by PS1 (Gezari et al. 2012). It had a similar light curve shape to PS1-11af (Figure 11), with similarly long-lived blue optical colors (Figure 6). However, the NUV to optical color was bluer for PS1-10jh, implying a minimum T_BB of 30,000 K. Gezari et al. (2012) argue that the intrinsic T_BB had to be even higher, with a minimum of ~50,000 K being necessary to supply sufficient ionizing photons for the observed He ii emission lines at early times. A modest amount of reddening (E(B − V) = 0.08 mag) could reconcile this with the data.

We investigate the possibility of extinction for PS1-11af and find that E(B − V) ≈ 0.2 mag is necessary for the best-fit T_BB of the Day +10 GALEX+PS1 SED to equal 30,000 K, and doubling that value pushes T_BB up to nearly 10⁵ K. However, such a large extinction would imply that L_BB is 1.6 × 10⁴⁶ erg s⁻¹, more than ten times the Eddington luminosity (L_Edd) for a 10⁷ M_☉ black hole. If we assume that the maximum permitted L_BB is ~3 L_Edd for M_bh ≈ 10⁷ M_☉ (the factor of three is to be conservative and allow for some uncertainty in the parameters), then we can set an upper limit on the reddening of E(B − V)  0.35 mag, which corresponds to an upper limit of T_BB  5.4 × 10⁴ K. While the best fit for the host galaxy SED has zero reddening (Section 3.1), we cannot exclude the possibility of some gas and dust local to the environment of PS1-11af in the nucleus. The broadband SED fits for Sw 1644+57 implied substantially more extinction for the transient than was derived for the host galaxy (Bloom et al. 2011; Levan et al. 2011; Burrows et al. 2011; Zauderer et al. 2011). We note that none of our galaxy-subtracted spectra exhibit any narrow absorption lines from intervening gas (e.g., Mg ii λ2800, Ca ii H+K, or Na i D), which argues against a large gas column along the line of sight.

The two optically selected SDSS TDEs of van Velzen et al. (2011a) were discovered on the decline, so there are only lower limits on the peak luminosities, but SDSS TDE2 appears to be most analogous to PS1-11af. It peaked at L_g > 4.1 × 10⁴³ erg s⁻¹ and had an average T_BB = 18,200 K, both of which are fairly similar to PS1-11af. van Velzen et al. (2011a) identified several distinctive characteristics of their two events compared to other transients and variable AGN in Stripe 82, including the extremely blue color and slow luminosity and color evolution. They parameterized the slow evolution using the somewhat unusual units of dln L_g/dt and dln T_BB/dt. We estimate L_g from λf_λ in r_P1 and T_BB from the BB fits to the PS1 g_P1r_P1i_P1z_P1 photometry above. We fit linear relationships to the logarithms of both of these quantities after maximum light and find dln L_g/dt = (−1.8 ± 0.2) × 10⁻² d⁻¹ and dln T_BB/dt = (−0.2 ± 2) × 10⁻³ d⁻¹, very similar to the values for both of the SDSS TDEs.

van Velzen et al. (2011a) also noted the unusual combination of the very blue color and its slow evolution for their objects. Our photometry does not extend sufficiently blue to cover the SDSS u' band, so we cannot directly compare our measurements to the observer-frame quantities used by van Velzen et al. (2011a). However, the higher redshift of PS1-11af allows us to compensate somewhat for this. We de-redshift our Day −5 and +24 spectra to z = 0.2 (similar to the SDSS TDEs) and use STSDAS/SYNPHOT²⁰ in IRAF to synthesize observer-frame colors. The average colors of PS1-11af from the two shifted spectra are u' − g' = −0.24 mag and g' − r' = −0.18 mag. We also fit a line to the g_P1 − z_P1 points (fairly close to u' − r' in the rest frame) from Figure 6 and find a slope of (0.1 ± 3) × 10⁻³ mag d⁻¹, consistent with no evolution. The combination of these colors and their stability over a long time baseline places PS1-11af in the same parts of parameter space as both of the SDSS TDEs in the diagrams of van Velzen et al. (2011a), and away from the SNe and AGN-like variables. This is further evidence that their objects are of a similar class as PS1-11af.

Cenko et al. (2012a) described the discovery of PTF 10iya, a fast evolving and UV bright nuclear flare. They modeled it as being the result of the early super-Eddington phase of accretion following the tidal disruption of a solar-type star by a ~10⁷ M_☉ black hole. That object was very different from PS1-11af, despite a similar T_BB. It had a somewhat higher peak luminosity (10⁴⁴ erg s⁻¹, depending on the extinction correction) but declined very rapidly (~0.3 mag d⁻¹) compared to PS1-11af. PTF 10iya also had a bright associated X-ray source, with L_X ≈ 10⁴⁴ erg s⁻¹. Unfortunately, we have no constraints on the high-energy emission from PS1-11af.

The two relativistic TDEs, Sw 1644+57 and Sw 2058+05, have much more limited optical data (apparently because of high extinction in the case of Sw 1644+57: Bloom et al. 2011; Zauderer et al. 2011). Sw 2058+05 did have a slowly evolving UV/optical light curve, but T_BB had a lower limit of 6 × 10⁴ K and L_BB  10⁴⁵ erg s⁻¹, indicating a significantly more luminous accretion event than for PS1-11af with a bluer SED (Cenko et al. 2012b).

Several UV-selected TDE candidates have been found in GALEX observations (Gezari et al. 2008, 2009b). These events had bluer SEDs than PS1-11af (α ≈ 1.1–1.4), and their inferred T_BB are correspondingly hotter. Single-temperature BB fits found T_BB in the range (4.4–12) × 10⁴ K (Gezari et al. 2009b), which is more comparable to the 10⁵ K expectations for accretion disks near the tidal radius. The light curves for these objects are also generally consistent with a t^−5/3 decline, although the lack of data points on the rise allows for some freedom in the light curve fits.

The TDE candidate D3-13 (Gezari et al. 2008) may be instructive for PS1-11af. A Chandra detection at late times required T_BB > 1.2 × 10⁵ K, while the UV/optical data at earlier times were better fit by a separate T_BB ~ 10⁴ K component in a two-temperature fit. The data were not taken simultaneously but may point to the existence of a hotter component that is not obvious in the optical data, even if it dominates the bolometric luminosity (Gezari et al. 2008).

We use the bolometric light curve from Figure 11 (the observed g_P1 light curve multiplied by a bolometric correction to match the L_BB from the GALEX+PS1 SED fit on Day +10) to constrain the accretion properties of PS1-11af. Third-order polynomial fits to the data around peak give a maximum L_bol of (8.5 ± 0.2) × 10⁴³ erg s⁻¹, where the errors are derived by Monte Carlo resamplings of the observed light curve. This translates to L_bol/L_Edd $\approx 0.07_{-0.03}^{+0.13}$. Similarly, we perform a trapezoidal integration of the light curve to find the minimum total radiated energy of (4.1 ± 0.1) × 10⁵⁰ erg between the first and last detections. If we assume a radiative efficiency factor, η ≈ 0.1, then the minimum necessary accreted mass onto the black hole is M ≈ 0.002 (0.1/η) M_☉. This is only a lower limit because it does not include the tail of the light curve at late times and, as we saw above, a small amount of extinction can dramatically increase the flux in the UV where the peak of the SED is, and we have few constraints on the SED shape. Still, the low value for L_bol (and hence the accreted mass) is suggestive of a scenario involving only a partial disruption of a star.

The observed range of R_BB is ~(5–12) × 10¹⁴ cm (Figure 8). These values are not straightforward to interpret in the TDE case because they were derived from the normalization of the BB fits by assuming a spherical emitting surface. The geometry of the TDE accretion flow could be quite different, with most models assuming that a disk dominates the UV/optical continuum unless reprocessing is invoked (Loeb & Ulmer 1997; Lodato & Rossi 2011; Strubbe & Quataert 2011; Guillochon et al. 2013). Nevertheless, these radii provide a useful scale for the system, as they correspond to 135–400 Schwarzschild radii. Even allowing for uncertainty in M_bh, they are far outside the expected tidal radius for a main sequence star. These radii are near the tidal radius for red giant stars, but TDEs with disruption radii that far from the central black hole are expected to have light curves with rise times of order a year, rather than a few weeks (MacLeod et al. 2012b).

We emphasize that these numbers have substantial systematic uncertainties. Most of the energy is emitted in the UV, where we only have the limited GALEX observations to constrain the SED and light curve, although T_BB determined from the PS1 observations alone is consistent with the fit to GALEX+PS1. Our assumption that the SED can be approximated as a single-temperature BB is definitely an oversimplification, as demonstrated by the UV excess of the galaxy-subtracted spectra relative to a BB (Figure 7). We know that some UV absorption features are present on Day +24, near the time of the GALEX observations, so if any others were present in the NUV bandpass, the UV flux might be suppressed and T_BB would be underestimated.

We have no information about any possible emission extending to higher energies. In the cases of PTF 10iya and D3-13, the X-ray emission carried a comparable or greater amount of energy to the UV/optical component without lying on an extrapolation of the low-energy SED (Cenko et al. 2012a; Gezari et al. 2008). If such emission were present in PS1-11af, it would be undetectable in our dataset. Our derived values for $\dot{M}_{\mathrm{acc}}$ are therefore best regarded as lower limits.

We now consider the shapes of the light curves to see if they are consistent with expectations from numerical modeling of stellar disruptions. We start with the estimated bolometric light curve from above and convert it to a mass accretion rate, $\dot{M}_{\mathrm{acc}}$, assuming that η = 0.1. The output $\dot{M}_{\mathrm{acc}}$ of numerical simulations are self similar, with the time and accretion rate variables being rescalable functions of M_bh and the mass and radius of the disrupted star, M and R.

We first fit to the simulations of Lodato et al. (2009), which were performed for the full disruption of a star at the tidal radius with a range of polytrope indices (γ; assuming an equation of state with P∝ρ^γ). We focus on models with γ = 5/3, appropriate for low-mass stars, because of the small accreted mass for PS1-11af. We find an excellent fit (top panel of Figure 12) for a disruption occurring 39 d prior to the peak of the light curve if we scale the time variable from the fiducial value by 0.73 ± 0.03. With the scalings from Lodato et al. (2009), this implies M_bh = (5.4 ± 0.5) × 10⁵ (M/M_☉)² (R/R_☉)⁻³ M_☉. Even for a representative low-mass main sequence star with M = 0.3 M_☉ (and R = 0.3 R_☉; Tout et al. 1996), M_bh is still ~1.8 × 10⁶ M_☉, well below our expectations from the host scaling arguments. Also, we had to apply a large vertical scaling factor to the plotted curve because $\dot{M}_{\mathrm{acc}}$ is so low. The integral under the curve implies a total accreted mass of only 0.003 M_☉. This is inconsistent with the full disruption of even a low-mass star, which was one of the assumptions of the Lodato et al. (2009) simulations.

Zoom In Zoom Out Reset image size

Therefore, we also examine partial disruption models (Guillochon & Ramirez-Ruiz 2013), which parameterize the ratio of the tidal radius to the pericenter distance where the (partial) disruption occurs as β. Guillochon & Ramirez-Ruiz (2013) give scaling relations for their families of curves in terms of several observables, but two of the parameters, the time from disruption to the peak and the asymptotic power-law index, depend on knowing the unobserved time of disruption. We estimate the mass lost by the star as 2 × M_acc ≈ 0.006 M_☉, from which we estimate β ≈ 0.57. We fit our light curve with the β = 0.55 model of Guillochon & Ramirez-Ruiz (2013) in the bottom panel of Figure 12 and can find good fits with M = 0.65 M_☉ and M_bh = 10⁶ M_☉. This is not intended to be a complete search of all parameter space, but instead it is intended to demonstrate that the shape of the light curve is compatible with the curve of $\dot{M}_{\mathrm{acc}}$(t) from a partial-disruption event.

The fits with both sets of models prefer smaller values for M_bh (~10⁶ M_☉) by about an order of magnitude than we expect based on the mass of the host galaxy. In each case, it is driven by the short rise time of the light curve because the time axis scales as M_bh^1/2. Our first detection of PS1-11af is on Day −14.5 at a flux level that is a factor of ~3–4 below the peak. By contrast, PS1-10jh had a rise time after the first detection of ~50 d (Figure 11; Gezari et al. 2012). This is not just a consequence of PS1-10jh being brighter and more easily detectable at fainter flux levels because it still took more than 30 d to rise the final two magnitudes to maximum light. Gezari et al. (2012) found that the same γ = 5/3 model of Lodato et al. (2009) required a time stretch factor of 1.38, compared to 0.73 for PS1-11af. Similarly, Guillochon et al. (2013) found a good fit to PS1-10jh with their partial disruption models with β = 0.87 and M_bh = 10⁷ M_☉. Their derived M_bh was somewhat higher than expected from the host stellar mass relationship, which they attribute to scatter in the M_bh–M_bulge relationship.

A major problem with the interpretation of both fits to the PS1-11af light curve is that the connection between $\dot{M}$ of stellar debris returning to pericenter and the light curve in any given observed band depends on the details of the hydrodynamics of the gas (e.g., whether an outflow forms) and the radiative transfer, along with their evolution through the event. We use the observed constancy of the colors to estimate $\dot{M}_{\mathrm{acc}}$ from the g_P1 light curve with a constant scaling factor; however, it is not clear that this is always justified and is certainly not expected in basic models.

Lodato & Rossi (2011) computed multiband light curves for the Lodato et al. (2009) models, including the effects of a wind (Strubbe & Quataert 2009). None of their models exceed νL_ν of ~few×10⁴² erg s⁻¹ in the optical band, an order of magnitude below that observed for PS1-11af, and most are closer to 10⁴¹ erg s⁻¹. This is a consequence of the models having hotter spectra than PS1-11af and emitting more of their radiation at higher energy. In addition, the model spectra evolve strongly with time, leading to light curve shapes in each band that are quite different from the mass return rates derived from the properties of the disrupted stars. Gezari et al. (2012) also found that PS1-10jh had a light curve shape that closely matched the shape of the disruption models of Lodato et al. (2009) and lacked the expected color evolution.

Guillochon et al. (2013) were able to fit the shapes of the light curves of PS1-10jh by including a reprocessing component to convert the accretion disk luminosity to a softer component with a roughly constant temperature, although their model invoked an unusually gray dust extinction law. They emphasize that pure accretion disk models cannot simultaneously satisfy the condition that the luminosity directly follows $\dot{M}_{\mathrm{acc}}$ and maintain a constant color. The origin of the reprocessing material is not understood. It could result from shocks at the disruption radius as the returning stellar debris interacts with itself or it could represent a version of the accretion disk winds seen in regular AGN (Murray et al. 1995). Reprocessing of some form has long been invoked to explain AGN SEDs, which also exhibit low disk temperatures relative to the expectations of naive thin disk models (Koratkar & Blaes 1999). See Lawrence (2012) for a recent review of this issue and some possible solutions.

Strubbe & Quataert (2011) predict the existence of a TDE outflow, but only as long as the accretion rate is super-Eddington. After maximum light, as the rate of mass return to pericenter drops, the optical depth in the wind drops and leaves the hotter disk more exposed (Lodato & Rossi 2011). T_BB for PS1-11af does not evolve in this fashion. For our inferred accretion rate for PS1-11af to approach Eddington would require invoking some UV extinction, an unobserved high-energy emission component (or at least more emission at shorter wavelengths than implied by the Wien tail of our single-temperature BB fits), or a smaller than expected M_bh. In an outflow scenario, the slight decrease in R_BB after maximum light could be explained by the decrease in $\dot{M}_{\mathrm{acc}}$ (and hence L_BB) providing less radiation pressure and a weaker wind. Strubbe & Quataert (2011) also include the effects of the unbound material from the disrupted star in their models, but the simulations of Guillochon et al. (2013) indicate that self-gravity confines the unbound material. It also has little effect on the SED or spectrum.

In summary, the shape of the light curve of PS1-11af can be acceptably fit by both full and partial tidal disruption models (Figure 12). However, the normalization to the low observed luminosity requires an accreted mass that is too low for a full disruption of a star. This inconsistency with the full disruption model leads us to favor a partial disruption scenario. However, if the bulk of the luminosity is emitted at higher energy, then the total accreted mass could be significantly higher. We note that the two models shown in Figure 12 have late-time decay rates that are in one case slower and in the other case faster than the canonical t^−5/3 value. This demonstrates the perils of attempting to match the observed late-time decay rate to theoretical expectations when the time of disruption is not observed, even for objects with light curves that are well sampled on the rise to maximum light.

Sw 1644+57 and Sw 2058+05 were both discovered due to a high-energy trigger and their long-lived, luminous X-ray counterparts were interpreted as the results of on-axis relativistic jets (Bloom et al. 2011; Levan et al. 2011; Burrows et al. 2011; Zauderer et al. 2011; Cenko et al. 2012b). We have no X-ray observations of PS1-11af, so we cannot exclude the possibility of such a high-energy counterpart here. However, any relativistic outflow should produce detectable radio emission, even if it is not oriented along our line of sight (Giannios & Metzger 2011). Most radio observations to date of other TDEs and TDE candidates have not detected any emission (Bower et al. 2013; van Velzen et al. 2013). However, there are one or two X-ray selected TDE candidates with late-time radio emission, potentially from off-axis jets (Bower et al. 2013).

Our three epochs of VLA non-detections strongly constrain the presence of any relativistic outflow. The redshift of PS1-11af is only slightly higher than that of Sw 1644+57, so an equivalently powerful relativistic jet would be easily detectable, even if it were viewed off-axis. As shown in Figure 13, the peak 5.8 GHz radio flux of Sw 1644+57 (Zauderer et al. 2011; Berger et al. 2012) is a factor of 100–300 above our non-detections of PS1-11af on similar timescales.

Zoom In Zoom Out Reset image size

Several models for the radio emission from jets produced by TDEs exist in the literature. One class invokes an analogy to gamma-ray burst (GRB) afterglows, with both reverse and forward shocks (Giannios & Metzger 2011; Metzger et al. 2012) from the decelerating blast wave potentially contributing to the observed radio emission. An alternative invokes internal shocks within the jet, by analogy to AGN jets (van Velzen et al. 2011b). A relativistic outflow with a kinetic energy of E_K becomes non-relativistic on a timescale of ~300 (E_K/10⁵² erg)^1/3n^−1/3 d, where n is the circumnuclear density, assumed to be uniform (e.g., Bower et al. 2013). On that timescale, comparable to that of our late-time observations of PS1-11af, the emission becomes more easily detectable by off-axis observers.

We take advantage of this fact to set limits on the presence of a relativistic jet by generating light curves using the GRB afterglow models produced by the BOXFIT code of van Eerten et al. (2012) rather than making detailed TDE jet models. The most important difference between jets produced by GRBs and by TDEs is that the former have a single impulsive episode of energy injection and the latter can have energy injection extending to late times (Berger et al. 2012; De Colle et al. 2012). However, we assume that this distinction primarily affects the shapes of the model light curves and is less important for predicting the timing and flux of the peak in the radio band, which mostly reflect the total energy in the jet and its orientation relative to our line of sight (van Eerten & MacFadyen 2012).

We limit our models to those with microphysical parameters fixed to the best-fit values from afterglow models for Sw 1644+57 (Berger et al. 2012; Zauderer et al. 2013). BOXFIT assumes a constant density medium that we set to n = 1 cm⁻³. We initially use E_K = 10⁵² erg to match the total energy in the jet measured for Sw 1644+57 for an assumed opening angle of θ_jet = 0.1 (Zauderer et al. 2013). Light curves observed 30° and 60° from the axis of such a jet are shown in Figure 13 and the peak fluxes clearly violate our upper limits for PS1-11af. If we scale the energy in the jet down by a factor of 10, the 60° off-axis case is only marginally consistent with the data. Also, a jet as powerful as Sw 1644+57 oriented in the plane of the sky (90° off-axis) could still be consistent with our limits because the radio peak is pushed to even later times (as expected; van Eerten & MacFadyen 2012). An unusually low density medium could also suppress the radio emission. We defer more detailed consideration of the parameter space excluded by our limits on off-axis jet production to future work.

Jet formation in TDEs could be strongly tied to the accretion rate relative to Eddington (Giannios & Metzger 2011; De Colle et al. 2012; Tchekhovskoy et al. 2013; van Velzen et al. 2013; Bower et al. 2013). By analogy with X-ray binaries, Tchekhovskoy et al. (2013) hypothesize that jets are features of either super-Eddington or strongly sub-Eddington phases. The jet in Sw 1644+57 can be modeled by assuming that energy injection ended when the accretion rate dropped below a threshold value (Zauderer et al. 2013; De Colle et al. 2012). One possibility for PS1-11af not forming a jet is that the peak accretion rate of ~0.07 $\dot{M}_{\mathrm{Edd}}$ simply never reached a sufficiently high value. If M_bh really is closer to 10⁶ M_☉, as derived from the light curve fits, or if there is a significant unobserved high-energy emission component, then $\dot{M}_{\mathrm{acc}}$ would be closer to the Eddington value and the lack of jet formation would have to be tied to some other parameter, such as the black hole spin or the magnetization of the stellar debris. Only ~10% of optically luminous quasars are radio loud (Kellermann et al. 1989), and a similar fraction could be applicable to TDEs (Bower et al. 2013).

We now return to the most novel aspect of PS1-11af compared to previously observed TDEs and TDE candidates, the transient broad UV absorption features. A zoom-in on the two absorption features is shown in Figure 14. We fit Gaussians to the two absorption features in the galaxy-subtracted Day +24 spectrum. The shortest-wavelength one has a centroid of 2470 Å and a FWHM of 10,100 ± 1200 km s⁻¹. The other one is centered at 2680 Å with a FWHM of 10,200 ± 400 km s⁻¹. The rest-frame equivalent widths are ~25 and 50 Å, respectively. No additive offset has been applied to the flux scale in Figure 14, so the zeropoint is appropriate for the spectra. The 2680 Å absorption has a minimum at about half of the interpolated continuum flux. Although the Day −5 LDSS spectrum becomes noisy very rapidly at shorter wavelengths due to the low instrument sensitivity in the blue, the 2680 Å absorption would be quite prominent if it were present.

Zoom In Zoom Out Reset image size

The redshift of PS1-11af is higher than most of the other TDEs, so the very few available spectra for other objects do not generally cover the wavelengths of these features, and we cannot determine whether they are present. For example, the spectra of PS1-10jh and SDSS TDE2 do not extend sufficiently far to the UV (Figure 10; Gezari et al. 2012; van Velzen et al. 2011a). An important exception is Sw 2058+05, which is at the much higher redshift of 1.1853 and has no obvious broad spectral features in the rest-frame UV despite high-quality spectra (Cenko et al. 2012b). The 2680 Å feature would be on the blue edge of the observed wavelength range for the first spectrum of PTF 10iya, but it does not appear to be present (Cenko et al. 2012a).

PS1-11af also lacks the optical emission lines seen in some other objects. PS1-10jh exhibited broad He ii λ4686 and λ3203 emission prior to maximum light, with FWHMs of 9,000 ± 700 km s⁻¹ (Gezari et al. 2012), and SDSS TDE2 had a broad Hα line after maximum light with a FWHM of 8000 km s⁻¹ (van Velzen et al. 2011a). None of these lines appear in any of our PS1-11af spectra at any epoch, and it is clear from Figure 10 that we would have easily been able to detect a He ii λ4686 emission line with the same equivalent width as that seen in PS1-10jh. Gezari et al. (2012) argued that the observed SED of PS1-10jh did not supply enough ionizing photons to explain the observed He ii line fluxes, and so invoked a modest amount of extinction and a higher intrinsic T_BB. PS1-11af is not as blue as PS1-10jh, so one possibility is simply that it has a lower intrinsic T_BB and cannot ionize sufficient helium for detectable emission. It is also perhaps not a coincidence that the FWHMs of the absorption features in PS1-11af are similar to the widths of the emission features in the other objects (and in AGN) and may indicate that the line-forming regions are at similar distances from the central black hole (in units normalized to M_bh).

We now consider possible identifications for these features. The red wing of the 2680 Å absorption rejoins the apparent continuum level near 2800 Å (Figure 14). The Mg ii λ2800 doublet is a very natural identification for this feature because it is strong in a wide variety of astronomical objects, and there are few strong isolated lines at nearby wavelengths. As stated in Section 5, the absorption minimum is blueshifted by 13,000 km s⁻¹, a reasonable value given the FWHM of the feature. If the other feature is blueshifted by a similar amount, it implies that the rest wavelength of the absorbing ion should be near 2578 Å. One possible candidate is Fe ii, which has strong resonance lines at λ2586 and λ2600. However, the feature at similar wavelengths seen in SLSNe such as our PS1-10bzj comparison has been identified as Si iii λ2542, following Quimby et al. (2011), and supported by detailed radiative transfer models (Dessart et al. 2012), although some Fe ii may contribute to the blend (Lunnan et al. 2013).

Strubbe & Quataert (2011) made predictions for spectral signatures of a wind produced by super-Eddington accretion at early times in TDEs. We inspect their output spectra, and the only line produced between 2500 and 3000 Å in any of their models is Mg ii. Their models with the Mg ii absorption also predict stronger Balmer and optical He ii absorption, which we do not observe. Their models assume a hot input spectrum (T_BB  10⁵ K) from the disk at the base of the wind leading to significantly hotter output spectra than we measure for PS1-11af and hence a higher degree of ionization in their spectra. We conclude that the Mg ii identification for the 2680 Å absorption is robust, but we regard the other one as more uncertain.

Broad UV absorptions are seen in ~10% of quasars, known as the BALQSOs (Weymann et al. 1991). Approximately 15% of these exhibit absorption in low-ionization lines, including the same Mg ii and Fe ii lines as those (possibly) present in PS1-11af (Voit et al. 1993). We show a composite spectrum of low-ionization BALQSOs (Brotherton et al. 2001) in Figure 14 as a comparison. This may be somewhat misleading, as the process of making a composite spectrum averages over discrete absorptions that are frequently narrower or detached from the rest wavelength. The typical shapes of the absorption troughs in BALQSOs (e.g., Voit et al. 1993; Hall et al. 2002) are generally quite different from the single broad absorption with a smooth profile for each line in PS1-11af. At lower luminosities, the intrinsic UV absorption lines in Seyfert 1 galaxies tend to be narrow and located at much lower velocities than the ~10⁴ km s⁻¹ we see here (Crenshaw et al. 1999).

Low-ionization BALQSOs are on average redder than normal quasars due to dust extinction in the absorbers (Sprayberry & Foltz 1992; Brotherton et al. 2001), but PS1-11af is instead bluer than normal quasar SEDs. While the high-ionization absorption lines in BALQSOs may be formed coincident with the BLR (Murray et al. 1995), recent photoionization work has determined that the low-ionization absorbers are located several kpc from the central black hole (e.g., Moe et al. 2009). We conclude that although the absorption features in PS1-11af potentially come from similar ions as those seen in low-ionization BALQSOs, the physical situation is very different.

Assuming a virial equilibrium with v² ≈ GM_bh/R and M_bh = 10⁷ M_☉, the expected typical velocities (v) for material located near our derived R_BB on Day +24 are 12,000 km s⁻¹. This is impressively close to the measured absorption blueshift of the Mg ii line, given the uncertainties in M_bh. There are also geometric uncertainties in the interpretation of R_BB in an accretion scenario and the factors of order unity involved in relating the absorption velocities to the virial velocity if they are formed in some sort of outflow with an asymptotic velocity that is a fraction of the escape speed. Still, this approximate equivalence, along with the fact that the line absorption is completely blueshifted from the rest wavelength, is evidence that the line formation region is in an outflow just outside of the continuum formation region.

Guillochon et al. (2013) argued that the lack of hydrogen lines in PS1-10jh can be explained by high ionization in the line-formation region. They analogize to reverberation mapping results from AGN (Blandford & McKee 1982; Peterson et al. 2004), which imply that the typical distances of Hα and Hβ emission from the central black hole for AGNs with the continuum luminosity of PS1-10jh are larger than the outer radius of the debris disk from the disrupted star. Such an explanation is harder to understand here if our line identifications are correct. It is true that our maximum values for R_BB are 1.2 × 10¹⁵ cm (~0.5 light-days), which is significantly closer than the typical Hβ lags of ~10 d for AGN with continuum luminosities similar to that of PS1-11af (Peterson et al. 2004). However, the Mg ii λ2800 emissivity of BLR clouds closely tracks that of Hβ under a wide variety of density, ionizing flux, and ionizing SED assumptions (Korista et al. 1997), and empirically, the FWHMs of the two emission lines are identical (McLure & Jarvis 2002). This implies that the formation regions for these two lines should be very similar. Our spectrum with the likely Mg ii absorption lacks any evidence of Hβ in emission or absorption (and does not extend sufficiently far to the red to include Hα). Furthermore, the Fe ii emission (at least in the optical) has been shown to originate in the outer parts of the BLR, farther than the Hβ region (Barth et al. 2013), although the only available reverberation lag measurement for the same resonant UV1 Fe ii multiplet that we likely see here suggests that it may form closer to the high-ionization C iv and Lyα lines (Maoz et al. 1993).

One possible way to understand the simultaneous appearance of Mg ii and the lack of Balmer lines starts from the observation that our spectrum with the line features lacks strong Mg ii emission as well (cf. the BALQSO composite in Figure 14). The balance between the emission and absorption parts of the profile depends on the optical depths and relative importance of scattering versus true absorption in a particular line, as well as the geometry and density profile of the emitting region. In pure scattering and a spherical geometry for an outflow, we might expect to see some SN-like P-Cygni emission as well as absorption if there is an outflow.

Strubbe & Quataert (2011) instead argued that the optical depths of resonance lines in the TDE outflow will be small, and in contrast to the line-driven disk wind models for BALQSOs (e.g., Murray et al. 1995), the absorption region in TDE outflows will form close to the continuum photosphere, which limits the geometric extent of the emission region and the equivalent width of any emission. Although the spectra of PS1-11af lack a clear quasar-like Mg ii emission line, there is weak evidence for some broad emission. In Figure 14, we fit power laws to the 3000–6000 Å continua and extrapolate them to the UV (dashed lines). On Day −5, the power law remains a decent fit at bluer wavelengths (with a little excess emission). However, on Day +24, there appears to be some excess relative to the power law near 2600 and 2800 Å, to the red of each of the absorption lines, with the continuum returning to match the best fit power law near 2400 Å. This is only weak evidence because it is sensitive to our assumption about the intrinsic continuum shape. If true, it is also possible that some of the UV excess relative to a BB seen in Figure 7 is actually a superposition of numerous weak emission features, as in the "Little Blue Bump" of quasars.

Observationally, some SNe have suppressed Balmer P-Cygni features at early times when the ejecta are hot, even when the ejecta are known to be hydrogen rich. The SLSN 2008es, which we used above as comparison for the color evolution of PS1-11af, is a good example. The Balmer lines did not become distinct until T_BB dropped below ~15,000 K, below the temperatures measured for PS1-11af (Miller et al. 2009; Gezari et al. 2009a). Even in normal SNe II at very early times, a similar effect is present when T_BB is near 2 × 10⁴ K and very steep density gradients in the outer ejecta lead to a very small line formation region relative to the continuum photosphere (Dessart et al. 2008), but such an explanation would not apply in a situation with deep absorptions, where the line formation region is likely extended. Nevertheless, perhaps it is possible for radiative transfer effects to suppress the hydrogen absorption below the expectations of Strubbe & Quataert (2011), although more detailed work is needed to verify this.