SPECTRAL EVOLUTION OF THE EXTRAORDINARY TYPE IIn SUPERNOVA 2006gy

We obtained visual-wavelength spectra of SN 2006gy at several times during the first year after explosion, as well as at one epoch more than 1 yr after discovery. Table 1 summarizes the observational parameters for each epoch. These were obtained at the Keck Observatory using the low-resolution imaging spectrometer (LRIS; Oke et al. 1995) and the Deep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003). We also obtained spectra using the Kast double spectrograph (Miller & Stone 1993) mounted on the Lick Observatory 3 m Shane telescope. The spectra at early times were all obtained with the long slit oriented at the parallactic angle (Filippenko 1982) to minimize losses caused by atmospheric dispersion, while later spectra when the SN faded had the slit passing through the nucleus of NGC 1260 (see Figure 1).

The spectra were reduced using standard techniques (e.g., Foley et al. 2003). Our spectra on days 36, 71, and 96 were already presented in Paper I, and additional details of the data-reduction steps can be found there. The spectra were corrected for atmospheric extinction (Bessell 1999; Matheson et al. 2000) using standard stars observed at an air mass similar to that of the SN and then flux calibrated using our photometry of SN 2006gy. When correcting the spectra for extinction, we adopt a host-galaxy extinction of A_R = 1.25 ± 0.25 mag (A_V = 1.67 mag; E(B − V) = 0.54 mag) for SN 2006gy in addition to the Galactic extinction of A_R = 0.43 mag (A_V = 0.57 mag; E(B − V) = 0.18 mag), and we use R = 3.1 and the reddening curve of Cardelli et al. (1989), as described in Paper I. As noted there, the Na i D absorption in the spectrum of SN 2006gy may suggest even higher reddening than we have assumed here, so our luminosity estimates for SN 2006gy are conservative. The observed and flux-calibrated spectra of SN 2006gy are shown in Figure 2. In all of the spectra presented here, the observed wavelengths have been corrected to the rest-frame wavelengths using redshift z = 0.0179 (see Paper I).

Zoom In Zoom Out Reset image size

Photometric and spectroscopic observations of SN 2006gy are challenging because the SN is located only ∼1'' from the bright galactic nucleus of NGC 1260 (Figure 1). This is especially true at late times when the SN has faded; during the main light-curve peak, SN 2006gy was much stronger than shown in Figure 1, dominating the surrounding flux (see Figure 1 in Paper I). There is a steep intensity gradient in the underlying galaxy light that includes a dust lane and velocity-dependent emission lines from H ii regions that follow the rotation curve of the galaxy (Figure 1; Paper I; Ofek et al. 2007; Smith et al. 2008b). To subtract host-galaxy light, we sampled emission along the slit immediately on either side of SN 2006gy (an example slit is shown in Figure 1). We verified that the sky-subtraction procedure did not artificially introduce false narrow absorption components due to oversubtraction; H, He i, and other lines show similar narrow blueshifted absorption profiles, even though only Hα and [N ii] are strong in the background, and the observed narrow absorption components weaken as the SN fades.

To check the uncertainty introduced by the difficult host-galaxy subtraction on such a strong intensity gradient near the nucleus, we compare (Figure 3) two relatively late-time spectra of SN 2006gy on days 177 and 209 to an extracted spectrum of the background galaxy at the same position. This background spectrum is from day 421 when the SN is no longer detected in our data. The flat continuum level here has been scaled to the background intensity of ∼2 × 10⁻¹⁶ erg s⁻¹ cm⁻² Å⁻¹ arcsec² measured in the WFPC2 image. At the shortest wavelengths, the background flux that would contaminate a ∼1'' point source is comparable to the SN flux at late times, so careful subtraction is important. Indeed, below λ ≈ 4500 Å, late-time spectra of the SN are nearly identical to the background, so we have some concern that even our background-subtracted spectra may still include substantial galaxy light in the blue.

Zoom In Zoom Out Reset image size

In order to evaluate the level of remaining host-galaxy contamination even after this baseline subtraction (which may be imperfect because of differing slit-width/seeing combinations), we experimented with subtracting additional galaxy light using the day 421 spectrum to represent the background at the SN position. The day 421 spectrum shows strong, narrow Hα and [N ii] emission lines from H ii regions. The most additional galaxy light one can subtract is the amount which causes the nebular [N ii] line to disappear in the SN spectrum; subtracting more than this would cause a spurious absorption feature at 6583 Å. Using this criterion, the days 177 and 209 spectra with the most possible host-galaxy light subtracted are shown in Figure 3 with gray tracings (orange in the online edition). The resulting spectra have weaker flux and steeper continua at the shortest wavelengths. Longward of 5000 Å, the contrast of some spectral features is stronger but the spectra are otherwise very similar. We suspect, however, that this may be an oversubtraction, because the method assumes that the SN has no intrinsic nebular emission from Hα and [N ii], whereas earlier spectra show variable fluxes of these lines.

Therefore, throughout this paper we adopt the original method, used consistently for all observed epochs, of subtracting the background by sampling emission along the slit on either side of the SN. The potentially oversubtracted spectra in Figure 3 are useful, however, to remind us of the level of uncertainty introduced by possible remaining galaxy light in the data. The differences introduced by the additional subtraction are not severe—they are comparable, for example, to the differences between LRIS and DEIMOS spectra at a given epoch (see Figure 4). If this contamination were as severe as the worst case represented in Figure 3, then the way that the additional subtraction of host-galaxy light would alter our analysis is that at late times after day 150 (and only at late times), we would derive a continuum temperature (Section 3.2) that is ∼5%–10% lower and an Hα equivalent width (Section 3.4.1) that is a few percent larger. Such differences are consistent with our estimated uncertainties, so it would not significantly affect our analysis. The subtraction issues discussed here would not alter the measurements of line flux and luminosity, and our conclusions do not rely on the flux at the shortest wavelengths.

Zoom In Zoom Out Reset image size

The inset of Figure 3 shows in detail the spectral region near Hα. In our late-time spectrum on day 421, we see no evidence for broad post-shock Hα emission that might arise from late-time CSM interaction, as we reported earlier (Smith et al. 2008b). The continuum underlying the narrow Hα and [N ii] emission lines is flat, in stark contrast to earlier epochs when broad Hα was present. Kawabata et al. (2009) reported broad Hα emission on day 394, estimated from a raised flux level in the gap between Hα and [N ii] λ6583. Figure 3 shows that at higher spectral resolution, this feature is not seen in our data on day 421. It therefore seems probable that the excess flux reported by Kawabata et al. (2009) may have been a blend of the wings of narrow Hα and [N ii] emission lines in their lower-resolution spectra, and may not have been attributable to the SN.

In general, our flux-calibrated spectra give reliable and self-consistent results. Compare, for example, the pair of spectra in Figures 2 and 4 on days 93 and 96, or the pair on days 122 and 125. These were obtained only a few days apart with different instruments on different telescopes, but the flux calibration, continuum slopes, and relative line strengths agree remarkably well except perhaps for the strengths of some of the narrowest emission or absorption features that are affected by the different instrument spectral resolutions.

We did encounter a potential problem in calibrating our spectra, however, and that was for the 2006 September 25 (day 36) spectrum obtained at Lick Observatory. This was our first spectrum of SN 2006gy, calibrated with standard stars that differed from those of all other epochs; more importantly, the blue-channel flat-field lamp on the Kast spectrograph was not working on this night, so we had to use a flat-field observation from a different observing run. These issues may compromise the reliability of the flux calibration and continuum slope on day 36, which is important for the resulting continuum temperature that is derived on this epoch. The red channel should not have been affected by the problematic blue-channel flat-field lamp, and this set of problems did not affect the remaining epochs of spectra. We therefore regard the blue continuum shape on day 36 with some caution when deriving physical quantities in our analysis. On the other hand, the strange continuum shape may be real and caused by unusual physical circumstances; it could result from higher extinction due to time-dependent circumstellar reddening, for example, since at that early time the optical depth was very high, as we discuss below. The dereddened spectra for all epochs are shown in Figure 4. We adopted a correction of E(B − V) = 0.723 mag (see Paper I), although for day 36 we also show an additional reddening correction of E(B − V) = 1.0 mag.

The overall spectral morphology of SN 2006gy can be characterized as a smooth blue continuum, convolved with a number of striking narrow and intermediate-width absorption and emission features (see Figure 4). Broad absorption features and broad P Cygni profiles typically seen in SN II-P atmospheres, with speeds of ≳10⁴ km s⁻¹, are absent in the spectrum throughout the main light-curve peak. The broadest lines in the spectrum have speeds of ±5000 km s⁻¹, where the line wings of Hα join the continuum.

During the slow rise to maximum and during peak luminosity, SN 2006gy is characterized by a smooth blue continuum with few emission features besides the strong Balmer lines. It has few absorption features other than the interstellar Ca ii and Na i lines. Following peak luminosity, SN 2006gy takes on a distinct spectral morphology that changes little during the decline phase for ∼100 days thereafter. The spectrum is punctuated by strong narrow emission and absorption features and intermediate-width absorption and emission. These produce interesting line-profile shapes with smooth red wings, but very sharp drops at v = 0 km s⁻¹ that are different from the smooth P Cygni profiles normally seen in SNe. The most striking examples of this are the Fe ii lines near 5000 Å (Figure 4).

The spectrum of SN 2006gy during this decline phase is unusual, but has some overlap with other SNe in various respects. The day 93 spectrum of SN 2006gy, which is exemplary of this phase, is compared with some other SN spectra in Figure 5.

Zoom In Zoom Out Reset image size

SN 2005gj. Before the discovery of SN 2006gy, SN 2005gj was the most luminous SN known. It was interpreted as a Type Ia explosion interacting with dense H-rich CSM, partly because its spectrum could be decomposed into a SN 1991T-like spectrum and a smooth polynomial (Aldering et al. 2006; Prieto et al. 2007). SN 2005gj is often grouped together with other luminous SNe IIn like SNe 2002ic, 1997cy, and 1999E (Germany et al. 2000; Turatto et al. 2000; Rigon et al. 2003; Hamuy et al. 2003; Wang et al. 2004; Deng et al. 2004; Wood-Vasey et al. 2004; Chugai et al. 2004; Chugai & Yungelson 2004; Kotak et al. 2004; Benetti et al. 2006; Chugai & Chevalier 2006) because their spectra are similar, although the connection to SNe Ia has been questioned (Benetti et al. 2006; Kotak et al. 2004; Trundle et al. 2008).

Agnoletto et al. (2009) found an overall remarkable similarity between the spectrum of SN 2006gy and this class of objects, but our comparison with SN 2005gj in Figure 5 contradicts that assessment because we find that at similar phases (93 versus 98 days) their spectra are very different. SN 2006gy does not show the broad absorption and undulations of the continuum seen in SN 2005gj (and its brethren SNe 2002ic, 1997cy, and 1999E), and its Hα emission is much weaker relative to the continuum.

SN 2006tf. SN 2006tf is the second most luminous Type IIn after SN 2006gy (Smith et al. 2008a). The similarity between spectra of SNe 2006gy and 2006tf (days 93 and 95, respectively) is better, as we noted in Paper I, although there is still some mismatch in blue spectral features. Relative to the continuum, Balmer-line emission is much stronger in SN 2006tf (note that the intensity scale in Figure 5 is logarithmic).

SN 2005gl. This SN IIn deserves special mention, because so far it is the only SN IIn for which a progenitor star has been detected in pre-explosion images (Gal-Yam & Leonard 2009). It is useful for comparison because it had a moderate luminosity, typical among SNe IIn (Gal-Yam et al. 2007). Early-time spectra were of Type IIn, but by about 2 months after explosion, the Type IIn signatures weakened and the spectrum began to look more like Type II-P. The day 34 spectrum that we show in Figure 5 resembles that of SN 2006tf, despite the much lower luminosity of SN 2005gl. Gal-Yam & Leonard (2009) inferred that the progenitor star was a L/L_☉ ≈ 10⁶ LBV-like star that, in the year or so before the SN, had suffered an outburst with $\dot{M} \approx 0.03$ M_☉ yr⁻¹. The event was therefore similar to the 1600 AD giant eruption of P Cygni (Smith & Hartigan 2006). The lower peak luminosity and relatively quick fading of CSM-interaction signatures in the spectrum of SN 2005gl are relevant to our study of SN 2006gy, because the precursor outburst that Gal-Yam & Leonard estimated, although extreme compared to normal winds, was tame compared to the precursor outburst needed for SN 2006gy.

SN 1994W. Perhaps the best overall similarity to the day 93 spectrum of SN 2006gy in Figure 5 is found with the spectrum of SN 1994W at day 89 (Chugai et al. 2004). The shape of the continuum and the amount of line blanketing at blue wavelengths show fairly good agreement. Both SNe share many of the same spectral features, although narrow lines are stronger relative to the continuum in SN 1994W. A more detailed comparison to our higher-resolution day 96 DEIMOS spectrum of SN 2006gy is shown in Figure 6, where it is clear that most of the narrow features are the same in both SNe, despite the different strengths. They are substantially narrower in SN 2006gy, with typical widths of 100–200 km s⁻¹ instead of 800–1000 km s⁻¹ in SN 1994W. Besides Balmer lines, most of these features are Fe ii (Chugai et al. 2004). Chugai et al. interpreted this emission/absorption spectrum as arising from a combination of an expanding cold dense shell and dense pre-shock gas expanding at 800–1000 km s⁻¹. Dessart et al. (2009), on the other hand, suggested that both components arise at the photosphere in the slowly expanding shell of SN 1994W, and that the broad wings arise from electron scattering. These aspects will resurface later.

Zoom In Zoom Out Reset image size

SN 1995G and SN 1998S. These are two well-studied and prototypical SNe IIn with lower luminosity, faster spectral evolution, and much faster decline rates than SN 2006gy (i.e., they behaved more like SN 2005gl). Their spectra at 90–100 days after explosion do not resemble spectra of SN 2006gy at a comparable epoch because they have already become optically thin by that point (see Leonard et al. 2000; Fassia et al. 2001; Pastorello et al. 2002). At earlier times when it was still optically thick, however, the day 36 spectrum of SN 1995G has a strong smooth continuum with narrow absorption/emission features similar to SN 2006gy, although its Balmer emission lines are stronger. Similarly, on day 25 SN 1998S also has a strong smooth continuum, but in this case the Balmer lines are weaker than in SN 2006gy. This gives a qualitative illustration that the strength of Hα is tied to the time evolution of the SN's optical depth. For SN 2006gy to maintain high optical depths for so long is remarkable. We will discuss the Hα line strength in more detail in Sections 3.3 and 3.4.

SN 1999em. This is a prototypical SN II-P spectrum in mid-plateau. As noted earlier, one would obviously not expect the recombination photosphere within a rapidly expanding SN envelope to match the Type IIn spectrum of SN 2006gy, and indeed the very broad and deep absorption features in SN 1999em are unlike narrow lines in SN 2006gy. However, we chose to display the spectrum of SN 1999em for the following reason: if we now dilute the photospheric spectrum of SN 1999em with a blackbody (second from bottom tracing in Figure 5), then we see a striking result. Ignoring discrepant expansion speeds, overall shapes of the two spectra match surprisingly well. This is especially true in the blue/near-UV regions, because this composite SN 1999em+blackbody matches the day 93 spectrum of SN 2006gy better than any SN IIn in Figure 5. This may suggest that the spectrum of SN 2006gy is generated by two components, such as a blackbody from ongoing CSM interaction combined with a recombination photosphere produced as radiation diffuses slowly out of an optically thick H shell (e.g., Smith & McCray 2007). Alternatively, it may mean that back radiation from the CSM interaction zone causes a muting of the line features via the "top-lighting" mechanism described by Branch et al. (2000). Diluting the observed spectrum in this way also provided the basis for comparing SN 2005gj to a SN Ia spectrum (Aldering et al. 2006; Prieto et al. 2007).

The dereddened visual-wavelength continuum shape can be approximated adequately with a blackbody function, with the caveat that there is substantial line-blanketing absorption at blue wavelengths, particularly severe below ∼4500 Å. No single temperature could account for the spectral shape both above and below 4500 Å, so in Figure 4 we show blackbody curves (orange) which approximate the λ>4500 Å continuum shape quite well. This drop in flux at short wavelengths is common in SN II-P atmospheres, as noted above, and in some SNe IIn. A few representative temperatures for the continuum of SN 2006gy are shown in Figure 4, but all are listed in Table 1.

The evolution of the apparent blackbody temperature is plotted in Figure 7 (middle panel), where the solid black line is a smooth functional approximation to the time-dependent temperature, since our spectra irregularly and sparsely sample the temporal evolution. The day 36 estimate of the temperature has large uncertainty, so we consider two different cases that bracket the early-time temperature evolution: lower temperatures around 10⁴ K (dashed curve) and higher temperatures near 15,000 K (solid). The lower temperature gives a better approximation of the continuum shape if we adopt the same reddening for all epochs (dashed day 36 spectrum in Figure 4). However, as shown in Figure 4, one can approximate the continuum shape somewhat better with a hotter 15,000 K blackbody if the day 36 spectrum is corrected for a larger reddening than the other epochs. This interpretation seems plausible given the high optical depths we expect in the inner CSM shell, as discussed later. Since the luminosity was still rising, CSM dust destroyed later at peak luminosity may still exist on day 36. The temperature was not much above 15,000 K, since the day 36 spectrum shows no sign of N iii and He ii features seen in the early phases of SN 1998S with temperatures above 20,000 K (Leonard et al. 2000).

Zoom In Zoom Out Reset image size

These two treatments of the temperature yield somewhat different bolometric corrections and emitting radii at early times (Figure 7), but have little impact on our overall interpretation. The dashed curve would suggest a temperature evolution different from that of other SNe IIn, whereas the solid curve with higher temperatures would make the temperature of SN 2006gy decline qualitatively similar to other SNe IIn, although slower. In either case, SN 2006gy takes much longer to cool, which is no doubt related to the extraordinary longevity of its peak luminosity phase. By day 150, the temperature settles to a floor at ∼6500 K and remains nearly constant thereafter. The fact that this temperature agrees with the same late-time value inferred for less-reddened SNe IIn (like SNe 2005gj and 1998S in Figure 7) could indicate that our assumed reddening value of E(B − V) = 0.72 mag is roughly correct. This temperature floor appears to be common among SNe IIn, whereas SNe II-P continue their decline to somewhat lower temperatures during their plateau phase.

These smooth curves approximating the temperature change of SN 2006gy are used to derive an approximate bolometric correction for the observed light curve in Figure 7 (top panel). The large filled circles correspond to the lower temperatures at early times (dashed curve), whereas the small circles show what the corrected luminosity would be with the higher assumed temperature (solid curve). Temperatures before day 110 lead to an increase in our estimate of the bolometric luminosity (solid circles) and a slight shift in the time of peak luminosity from 70 days to about 60 days after explosion. This bolometric correction affects the total radiated energy as well: in Paper I we found that the radiated energy integrated over the first ∼200 days was 1.6 × 10⁵¹ erg. With the corrected L_Bol light curve in Figure 7, this estimate rises to roughly (2.3–2.5) ×10⁵¹ erg (the uncertainty is due to the temperature evolution at early times).

With these bolometric luminosity and temperature curves, we also derive the effective blackbody radius, R²_BB = L/(4πσT⁴), plotted in the bottom panel of Figure 7. Again, the large solid points with error bars are for the lower temperatures at early times, and the small points show the derived radius for the higher temperatures. Differences between these two are minor, and do not much affect our discussion. The effective emitting radius of SN 2006gy behaves in a manner qualitatively similar to those of other SNe IIn in that R_BB increases steadily for a time, reaching a peak and then decreasing again. For SN 2006gy, this turnover occurred much later than in any other SN—R_BB turned over around day 115, whereas it occurred much earlier at day ∼50 for both SNe 1998S and 2005gj (Fassia et al. 2000; Prieto et al. 2007), and at day 70–80 in SN 2006tf (Smith et al. 2008a). In all cases, the turnover of R_BB was well after the time of peak luminosity.

Presumably this peak represents a critical transition point when the shell starts to become optically thin. In the case of SNe II-P, this occurs as the H-recombination photosphere recedes backward through the geometrically thick expanding H envelope of the star, marking a clear division in radius between the inner optically thick ionized ejecta and the outer recombined transparent ejecta. In SNe IIn, however, one expects the post-shock shell to form a geometrically very thin layer (a "cold dense shell," or CDS; see Chugai et al. 2004), because the post-shock layer collapses due to radiative cooling. In that case, the turnover and subsequent decrease of R_BB is misleading, because R_BB decreases more than the expected shell thickness, which should be much less than 10% of R_shell. Also, apparent velocities measured in the Hα line do not decrease during this time.

Motivated by similar observations of the luminous SN 2006tf, we (Smith et al. 2008a) explained this behavior in SNe IIn as follows: the derived value of R_BB is not the true radius of the emitting surface in the cold dense shell, but rather, R²_BB = ζR²_shell. The true physical radius of the shell, R_shell continues to increase with time, but ζ is a dilution factor representing the fractional geometric covering area of the shell (ζ ⩽ 1, decreasing with time). The post-shock shell is able to continue producing continuum radiation with a constant effective temperature of 6000–6500 K, despite its reduced effective opacity, because it is clumpy; small clumps remain optically thick longer, whereas less dense interclump regions gradually become transparent. One expects the CDS to be severely clumped because of Rayleigh–Taylor instabilities at the contact discontinuity. Limb-brightening effects may also be important in determining ζ.

The rate at which R_BB increases with time during early phases before this transition occurs (between days 40 and 120) seems to roughly match an expansion speed of ∼5200 km s⁻¹ (solid line in Figure 7). This speed is taken from the broadest Gaussian matched to the Hα emission profiles, which seem to imply constant expansion speeds, as we discuss below in Section 3.3. Hα continues to indicate roughly the same expansion speeds after the turnover in R_BB at day 115, when one expects that high optical depth effects like electron scattering wings will be less influential. The extrapolated radius around day 25 of ≲3 × 10¹⁵ cm or ≲200 AU roughly matches the expected initial radius of the highly opaque CSM shell inferred by Smith & McCray (2007), who claimed that it may have taken a few weeks for the SN to traverse this shell.

The evolution of the broad/intermediate-width component of Hα is a key diagnostic of the CSM interaction in SNe IIn, and will be discussed in detail in following sections. We show a time series of the Hα profile of SN 2006gy in Figure 8, and we display high-resolution line profiles at several epochs in Figure 9.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Caveats to keep in mind for the discussion below are the following. Our first three spectra on days 36, 65, and 71 were obtained with the Kast spectrograph at Lick Observatory, and have lower spectral resolution than the Keck/DEIMOS and LRIS data. This alters the appearance of the narrow components and the transition between the narrow and broad components on those dates in Figure 8. The red side of the Hα line is affected by [N ii] λ6583 emission to varying degrees, and the blue side by [N ii] λ6548 to a lesser degree; the positions of these lines relative to the Hα systemic velocity are marked in Figures 8 and 9. The Hα profile is also affected by weaker broad and narrow lines of He i and Fe ii, as noted in Paper I, whereas the blue wing of the line is depressed by varying amounts of broad blueshifted Hα absorption. For all profiles in Figure 8, we have normalized to the smooth continuum level (see Figure 4) and scaled the emission-line strength in order to compare profile shapes. We do not plot the Hα profiles on days 92, 93, 122, 127, 153, 158, and 177, because they are redundant for our purposes, superseded by higher-resolution DEIMOS data obtained within a few days.

For comparison, we plot each Hα profile atop a symmetric Gaussian, shown in gray in Figure 8 and in magenta in Figure 9. This symmetric profile is the sum of two Gaussians with equal intensity and FWHM values of 5200 km s⁻¹ and 1800 km s⁻¹. It is not necessarily the best fit to the profile on any specific date, but it gives an adequate match to the red wing for most epochs except as noted below. In addition, for the right-hand panel in Figure 8 and for the lower set of tracings in Figure 9, we show the residual profile after subtracting away this symmetric Gaussian in order to emphasize changes in the broad blueshifted Hα absorption.

3.3.1. The Pre-Maximum Hα Profile

Our earliest spectrum—the only one obtained while SN 2006gy was still on its slow rise to maximum luminosity—was obtained on day 36. This is an unfortunate consequence of the fact that the SN was not discovered until about 29 days past our adopted explosion date. (Upper limits and photometry before that time were extracted from pre-discovery data.) Also, SN 2006gy was initially thought to be an active galactic nucleus; its peculiar SN nature was not realized until relatively late (see Paper I and references therein).

As discussed below, the broad Gaussian curve in Figure 8 gives an adequate match to the red wing of Hα for most epochs, with a clear exception being the first spectrum on day 36. The broadest component on day 36 is substantially narrower than the Gaussian curve; consequently, subtraction residuals show a broad deficit in the redshifted wings.

Instead, the broad Hα wings on day 36 are better matched by a single Lorentzian profile, with FWHM = 1350 km s⁻¹ and with its centroid redshifted by +150 km s⁻¹ (Figure 10). This is not true for any epoch after day 90, when a single Lorentzian that fits the broad line wings and core would significantly underestimate the wings at ±1000–3000 km s⁻¹, and would require a net redshift of +450 km s⁻¹ (an example of such a profile plotted with the day 125 spectrum is shown in Figure 10). In addition to the broad Lorentzian, the line has an unresolved narrow Gaussian component that contributes ∼7% of the total Hα luminosity.

Zoom In Zoom Out Reset image size

A similar result was found for the Hα profile in early spectra of SN 1998S, where Leonard et al. (2000) showed that on day 5 it was well fitted by a broad Lorentzian plus an unresolved component that contributed 20%–25% of the line luminosity. Chugai (2001) proposed that this profile arises from an intrinsically narrow Hα line that is broadened by collisions with thermal electrons in an opaque circumstellar shell outside the photosphere. The Lorentzian wings arise from multiple electron scattering, whereas the unresolved narrow component is direct emission. A similar model may explain the early-time Hα emission from SN 2006gy. In this interpretation, the net +150 km s⁻¹ redshift of the Lorentzian centroid corresponds to the expansion speed of the CSM at these inner radii (see Chugai 2001). The scattering envelope's optical depth outside the photosphere, τ, is given roughly by U = (1 − e^−τ)/τ, where U is the ratio of the unscattered narrow component to the total line luminosity. For SN 2006gy on day 36 we measure U ≈ 0.07, yielding τ ≈ 15. The pre-SN envelope is evidently highly opaque.

The FWHM of this Lorentzian component on day 36 is only 1350 km s⁻¹, which is much slower than the necessary expansion speed of ∼5000 km s⁻¹ corresponding to d(R_BB)/dt in Figure 7. If anything, the true expansion speed of the photosphere should have been even faster before day 36. Thus, at this early phase, the line wings cannot be caused by electron scattering within the rapidly expanding CDS itself, as suggested for SN 1994W by Dessart et al. (2009), although this effect may play a role at later epochs (see below). Since no broader components are seen at early times, we must conclude that the Hα profile on day 36 does not trace the kinematics of the SN. Rather, it seems likely that the opaque photosphere occurs outside the forward shock, as a radiative precursor ionizes the relatively slow and extremely dense pre-shock CSM. This opaque ionized region hides the underlying SN kinematics from our view, so that the broad line profiles at early times are due entirely to Thomson scattering of narrow lines formed in the CSM, whereas the rising luminosity and increase in R_BB at early times is due entirely to a rapidly expanding ionization front. The dereddened Hα/Hβ ratio of ∼2.9 on day 36 (Table 1) seems consistent with this interpretation. This radiation cannot escape directly from the post-shock zone, but must diffuse out through the opaque envelope, as suggested by Smith & McCray (2007) to explain the slow rise to maximum of SN 2006gy.

Altogether, the spectrum of SN 2006gy in the time period when it was still slowly rising to maximum luminosity can be characterized by a blackbody photosphere expanding within an opaque electron-scattering circumstellar shell. This was also the case for SN 1998S, but only at the very earliest times up to day 5. The much longer duration of this phase in SN 2006gy is yet another testament to the extraordinarily dense and extended pre-SN envelope. It is unfortunate that additional spectra were not obtained during this early phase.

3.3.2. Nearly Constant-Velocity Line Wings

3.3.3. Broad Blueshifted Absorption

The measured Hα equivalent width and line luminosity are shown in Figures 11–13. The measured Hα equivalent width has several potential sources of error.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

First, we are comparing data obtained with different telescopes, instruments, and (most importantly) slit widths. This is not a severe problem at early times when the SN is bright, but it can be problematic at late phases. Figure 11 shows the equivalent width plotted with different symbols corresponding to the three spectrographs we used. Spectra obtained at Lick Observatory have the largest slit width, so these equivalent width measurements are likely to have the largest systematic error because of the difficulty in subtracting background light. Overall, we found that the LRIS and DEIMOS measurements were very consistent, whereas the Lick measurements at comparable epochs showed larger variation. For this reason, we adjusted the equivalent widths of the Lick measurements to correct for partial contamination by 10⁻¹⁵ erg s⁻¹ cm⁻² Å⁻¹ of galaxy continuum light in the slit, which brought the late-time equivalent width into agreement with the LRIS and DEIMOS values. This is justified by the similar line luminosities at those times, which are not altered by the continuum level.

Second, the Hα profile is complex, with several adjacent overlapping line features that change with time. For example, at late times, the red wing of Hα descends into a broad absorption feature at 6700–7000 Å, which alters the underlying continuum level. The blue wing is plagued by broad P Cygni absorption and by Fe ii and Si ii lines. These complexities are exacerbated by the different spectral resolutions of different instruments. We aimed to consistently measure the Hα equivalent width and flux over a wavelength range of 6400–6700 Å, although the effects noted above introduce uncertainty in the underlying continuum level at 5%–10%. Equivalent width and flux measurements listed in Table 1 reflect the total Hα line, including intermediate-width and narrow components. We made no correction for Fe ii and He i lines included in the sampled wavelength range.

3.4.1. Hα Equivalent Width

3.4.2. Hα Line Luminosity

3.4.3. The Hα/Hβ Line Ratio

In addition to the broad Hα component discussed above, SN 2006gy also has narrow Hα emission from the photoionized but unshocked CSM. In Paper I, we presented a high-resolution spectrum of Hα on day 96, which revealed a narrow blueshifted P Cygni absorption line as well. This component indicated outflow speeds of ∼260 km s⁻¹ in the pre-shock CSM. If this is attributed to the wind speed of the progenitor star, it suggests an escape velocity comparable to that expected in blue supergiants. If it is associated with an explosive pre-SN ejection, it is reminiscent of speeds seen from LBV eruptions (Paper I). It is much faster than speeds for a red supergiant (RSG), and much slower than for a Wolf–Rayet star.

Figure 14 shows the narrow Hα absorption and emission feature in SN 2006gy at several epochs as seen in our high-resolution DEIMOS spectra. Two qualitative points are evident from this plot.

• 1.  
  The strength of the blueshifted absorption weakens relative to the narrow emission component. On day 96 the two are comparable, implying that it may be a pure scattering P Cygni profile, but as time proceeds, this narrow blueshifted absorption weakens.
• 2.  
  The velocities of the absorption minimum and the blue edge of the narrow absorption feature shift to faster speeds at later times. The speeds of the absorption minimum are roughly −135, −157, −163, and −175 (±5) km s⁻¹ on days 96, 125, 154, and 179, respectively. The blue-edge velocities are less clear, but are roughly 280(±15), 330(±20), 355(±20), and 440(±35) km s⁻¹ on these same days. We define the blue-edge velocity as the point where the intensity turns over and begins to drop down the slope of the broad component, as shown in Figure 15. It is complicated by the presence of [N ii] λ6548 and by noise or complex structure in the data, so these values are approximate. Nevertheless, it is clear that the velocity changes with time in a systematic way.⁵

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The changing depth of the narrow absorption probably results from the simple fact that as the blast-wave radius increases with time and sweeps through the CSM, the total column density of the pre-shock gas decreases. This occurs because the immediate pre-shock density should drop with increasing radius, and because increasing amounts of CSM are swept up behind the shock.

The increasing speed seen in the narrow blueshifted absorption components (Figure 15) is interesting, as it directly samples the velocity profile of the pre-SN mass loss. We can use these observed velocities to reconstruct this pre-SN velocity profile and thereby infer the age of the CSM. If we assume that the narrow absorption at each epoch traces cool pre-shock material just outside the forward shock at that time, we can derive its approximate radius away from the progenitor by using the radius given in the bottom panel in Figure 7. (We also assume that the blast wave continues to expand at ∼5200 km s⁻¹ after day 110 when the derived value of R_BB decreases due to effective optical depth, as explained in Section 3.2.) Calculating the radius in this way, we plot corresponding values of R and V in Figure 16.

Zoom In Zoom Out Reset image size

The blue-edge velocities are probably the best indicator of the outflow speed of pre-shock gas along the line of sight at each epoch, whereas the absorption minimum probably traces velocities with some projection from the line of sight closer to the limb of the shell. Figure 16 demonstrates that the velocity of pre-shock gas along the line of sight increases with radius. We also plot the CSM speed of 150 km s⁻¹ inferred from the shifted centroid of the Lorentzian fit to the day 36 Hα profile, assuming that it is caused by electron scattering (see Section 3.3.1). For comparison in Figure 16, we also plot dashed lines for the velocity profile one expects for simple Hubble laws (V ∝ R) for CSM shell ejections that occurred 6, 7, 8, 9, and 10 yr prior to the SN explosion. Although the velocities do not perfectly match a linear increasing speed with increasing radius, they are consistent with such a trend within the uncertainties. The age of the CSM shell appears to be 7.8 ± 1 yr before explosion, if the expansion has been roughly constant.

There are several interesting implications. A Hubble law in the CSM velocity profile means that the observed speeds are not the consequence of acceleration of the pre-shock gas by the radiation force of the SN luminosity itself; that mechanism is expected to cause a negative velocity gradient (e.g., Chugai et al. 2002), which is the opposite of the observed trend. Perhaps the pre-shock CSM speeds of other SNe IIn, such as 400, 700, and 1000 km s⁻¹ around SNe 1998S, 1995G, and 1994W, respectively (Chugai et al. 2002, 2004; Chugai & Danziger 2003), may reflect their progenitors' ejection speeds, rather than radiative acceleration by the SN.

If the CSM speed really is tied to the physics of the progenitor's mass ejection, rather than having been radiatively accelerated afterward by the SN, it implicates a progenitor object with an escape speed significantly faster than a large RSG, as we discussed in Paper I. Instead, the observed Hubble-like velocity profile points to a relatively sudden episode of mass ejection with speeds of several hundred km s⁻¹. Regardless of the cause, this is similar to the range of velocities and Hubble-like profiles observed in the ejecta of massive LBVs such as η Carinae (Smith 2006a). In Paper I, we noted speculative connections between the precursor event of SN 2006gy and the 1843 eruption of η Carinae based on the H-rich composition, total energy of ∼10^49.7 erg, ejected mass of order ∼10 M_☉ (Smith et al. 2003), and the similar ejecta speed. The Hubble-like velocity laws seen in both objects reinforce this comparison. It is also a feature that is explicitly predicted in the model of Woosley et al. (2007), where the pre-SN mass loss is caused by a ∼10⁵⁰ erg pulsation or explosion triggered by the pair instability. The observed delay time of 7–9 yr between the precursor and SN 2006gy is in good agreement with the 6.8 yr delay predicted in Woosley et al. 's model. The Hubble-like profile is not consistent with a dense but steady mass-loss wind, indicating that the progenitor star was decidedly unstable leading up to core collapse. For SN 2006gy, the precursor event ejected ∼19 M_☉ at speeds of 200–500 km s⁻¹, with a corresponding kinetic energy of >10⁴⁹ erg.

The fact that the systematically increasing CSM velocity is seen in absorption means that it corresponds to material along our line of sight. The range of velocities is therefore not likely to be caused by asymmetry in the CSM, although by no means is this an argument that the CSM is spherical. We hope to investigate possible effects of asymmetry in a future paper discussing our spectropolarimetry of SN 2006gy. Asymmetry in the CSM is a potential explanation for the smooth, slowly evolving light curve (van Marle et al. 2009b).

Figure 17 gives a detailed view of our high-resolution spectra of SN 2006gy at visual wavelengths from the blue to Hα, emphasizing features that make the spectrum of SN 2006gy so distinct. The most remarkable characteristic of the spectrum is the presence of line profiles with both broad and narrow emission and absorption components, leading in some cases to profiles with extremely sharp edges that separate redshifted emission from blueshifted absorption. As noted earlier (Figure 5), narrow absorption features in SN 2006gy find their closest comparison with the spectra of SN 1994W and SN 1995G (Sollerman et al. 1998; Chugai & Danziger 2003; Chugai et al. 2004). However, the dual nature of the SN 2006gy profiles—with separate broad and narrow components—was not seen in those two SNe.

Zoom In Zoom Out Reset image size

A key point to recognize in Figure 17 is that after first appearing in the day 96 spectrum (and the days 92 and 93 spectra in Figure 4), the narrowest absorption features become systematically weaker with time. These narrow absorption features at velocities of roughly −200 km s⁻¹ are formed in the pre-shock CSM. Their progressive weakening is not an effect of increasing or decreasing ionization with time in the pre-shock CSM, since the same trend is seen in absorption of both He i and Fe ii lines.⁶

This effect signals that the column density of pre-shock absorbing material is decreasing with time as the forward shock approaches the outer boundary of a dense shell. The weakening of the narrow absorption features occurs from days 90 to 200; many of the narrow features are barely discernible on day 179. Recall (Section 3.4.2) that the total line luminosity of Hα begins to drop around day 150 and plummets severely by day 209 (Figure 13). The Hα emission line is produced primarily in the post-shock gas, which is becoming optically thin at these late phases, so it directly traces the strength of the CSM interaction. The fact that both this post-shock Hα emission and the pre-shock narrow absorption features weaken at the same time confirms the notion that by day 200, the blast wave of SN 2006gy is overtaking the outer boundary of a shell that was probably ejected ∼8 yr before core collapse.

Fe ii λ5018 and λ5169. Figure 18 shows the profiles of Fe ii λ5018 observed at high resolution. This is a strong line representative of many other features in the spectrum. The red wing of its smooth broad emission component indicates velocities similar to those of Hα, suggesting that it is formed in a similar region of the swept-up CDS. It is not present in the early-time spectrum before and during peak luminosity, when we expect that the photosphere is ahead of the forward shock, but broad Fe ii λ5018 emission is present at all times after the post-shock gas becomes optically thin. Figure 18 demonstrates the extremely sharp transition from redshifted emission to blueshifted absorption that is seen in many lines in the spectrum of SN 2006gy. The gray curve in Figure 18 is a symmetric Gaussian matching the red emission wing, showing that Fe iiλ5018 suffers from significant broad blueshifted self-absorption out to −4000 km s⁻¹, as seen in Hα. This broad absorption remains constant with time. Fe iiλ5169 has a similar broad absorption feature, but lacks the broad emission.

Zoom In Zoom Out Reset image size

Relative to the broad-emission strength, the narrow absorption from the pre-shock CSM weakens with time as noted above. Its equivalent width decreases by more than a factor of 2 from day 96 to 179. The velocity of the pre-shock gas causing the narrow Fe iiλ5018 absorption does not show as clear an increase with time as had been seen in Hα, but the day 96 spectrum is noisy. Fe ii λ5169 has a narrow absorption feature similar to that of Fe ii λ5018 (Figure 18), so we infer that the Fe ii λ5018 line is not severely affected by He i λ5015. The narrow Fe ii absorption indicates speeds in the CSM of 200–600 km s⁻¹, consistent with Hα.

Si ii λλ6347,6371. Figure 19 shows the profiles of the Si ii λλ6347,6371 doublet observed at high resolution. These lines are of particular interest since they are a defining characteristic of SNe Ia, and the presence of this absorption feature in SN 2006gy might imply a connection to objects like SN 2002ic and 2005gj, although as noted elsewhere (Ofek et al. 2007; Paper I), these lines are seen in some core-collapse SNe.

Zoom In Zoom Out Reset image size

The Si ii λλ6347, 6371 doublet exhibits narrow absorption that weakens with time, in agreement with several other absorption lines formed in the pre-shock CSM. The doublet shows no narrow or broad emission.

More interesting, perhaps, is the broad blueshifted Si ii absorption in Figure 19. The shape of the blue wing is compromised by a blend with Fe ii, but it is clear that significant broad blueshifted absorption of Si ii is present on day 96, out to velocities of roughly −4000 km s⁻¹. This absorption weakens with time, perhaps due to ionization effects, while it is weak or absent in early-time spectra. The speeds traced by this blueshifted Si ii are far too slow to arise in a normal SN Ia atmosphere, where speeds well above 10⁴ km s⁻¹ are seen, but velocities as fast as 4000 km s⁻¹ are entirely consistent with the expanding post-shock shell. Similar broad Si ii absorption was seen in SN 1998S (Fassia et al. 2001). We conclude that the broad Si ii absorption in SN 2006gy is not formed in the photosphere of the underlying SN, but is instead in the post-shock shell, and therefore provides no connection to a putative SN Ia.

Na i D and He i λ5876. The most striking features associated with the Na i line in Figure 20 are the four narrow absorption features that correspond to the two pairs of Na i absorption lines formed in the Milky Way ISM and in the ISM of NGC 1260. These features do not change much with time when observed at consistent resolution, and support the large values of Milky Way and host-galaxy reddening that we inferred in Paper I. Figure 20 also reveals narrow absorption and emission from He i λ5876 (and λ7065), consistent with other narrow He i features in the spectrum.

Zoom In Zoom Out Reset image size

The identification of the broad emission in the top panel of Figure 20 is not obvious. Due to the proximity in wavelength of Na i D and He i λ5876, it could be either or both at different times. It follows the same trend of increasing equivalent width of Hα emission over the same time period. Since Na i is expected to be strong under similar conditions in SN II spectra (e.g., Dessart & Hillier 2008) and because similar increases in strength are not seen in emission components of other He i lines (like He i λ7065; bottom of Figure 20), we surmise that the broad emission component is Na i.

The nature of the broad absorption in Figure 20 is more perplexing. This absorption strengthens with time, whereas Fe ii λ5018 stays roughly constant and Si ii decreases in strength with time. Unlike any broad absorption features discussed so far, it reaches to velocities as fast as −8000 km s⁻¹ on days 154 and 179. The broad absorption blueward of He i λ7065 behaves very differently in velocity and time (Figure 20, bottom) from that of He i λ5876.

The higher-speed absorption near Na i may be due to a second absorbing species that we refer to here as component 2 (see Figure 20). The relative contribution of component 2 changes with time; it is initially absent and then becomes prominent on days 154 and 179, forming an additional dip at velocities of −5000 to −8000 km s⁻¹ that was absent before day 154. It seems likely that this additional absorption is due to a superposed line of another species, since (a) the strength of component 2 changes differently from the rest of the broad absorption feature, (b) it is not seen in any other broad absorption line, and (c) fast SN ejecta with speeds of ∼8000 km s⁻¹ should have already caught up with the reverse shock by day 80 (see Section 4.3). It may, however, be related to other broad absorption features at red wavelengths that strengthen at the same time, like the broad absorption blueward of He i λ7065.

We obtained six epochs of high signal-to-noise ratio (S/N) spectra with LRIS on Keck (plus a last epoch on day 237 with lower S/N; see Table 1). These are useful for investigating the evolution of the red-wavelength spectrum of SN 2006gy (Figure 21), not sampled by our DEIMOS data. The day 71 spectrum is included in Figure 21, representative of our three early epochs of Lick/Kast spectra on days 36, 65, and 71, which exhibit essentially featureless blackbody continua at red wavelengths with perhaps some weak absorption in the Ca ii near-IR triplet and weak emission of He i λ7065.

Zoom In Zoom Out Reset image size

As shown by our sequence of LRIS spectra in Figure 21, the post-maximum red-wavelength spectrum evolves with time. Prominent emission features are He i λ7065, a blend of several Fe ii emission lines, O i λ7774, and the Ca ii near-IR triplet. The last of these is particularly interesting; it becomes nearly as strong as Hα, suggesting that it may be a significant source of post-shock cooling in SNe IIn. A possible blend of [O ii] and [Ca ii] near 7325 Å is also present, although we cannot be confident of the identification of these forbidden lines because they overlap with several Fe ii lines. The spectra in Figure 21 are roughly consistent with previous reports, but have better temporal sampling and wavelength coverage. Agnoletto et al. (2009) noted that the Ca ii triplet was present in their spectrum on day 174; our spectra show that this was after the time when this line was strongest and was already in a steady decline. Spectra on days 127 and 157 presented by Kawabata et al. (2009) showed He i λ7065 and a few of the Fe ii emission lines out to ∼7400 Å, although in our spectra near those dates, these features appear stronger relative to the continuum. Lines in our red spectra during days 100–240 do not correspond well with those in later spectra on day 394 identified by Kawabata et al. (2009).

Our spectra reveal that the flux of these features changes notably with time. In Figure 22, we compare the temporal evolution of the integrated luminosity of the Fe ii blend and that of Ca ii λ8662 with the Hα luminosity (from Figure 13). To represent the behavior of the Fe ii blend while avoiding the strong O i λ7774 absorption, we took the flux above the continuum as it appears in Figure 21 measured over ∼7130–7620 Å. For Ca ii λ8662, we measured the line strength integrating over roughly −2,000 to +4,000 km s⁻¹. These emission features are not seen in early-time spectra, so for those epochs we give conservative upper limits.

Zoom In Zoom Out Reset image size

Figure 22 shows a very interesting trend. The Fe ii and Ca ii emission track the behavior of the Hα luminosity almost exactly after day 90, but at earlier times they are relatively much fainter than Hα. The likely implication is that the Fe ii lines and the Ca ii triplet trace emission directly from the post-shock cooling shell that is the product of CSM interaction, which is invisible at early phases because it is behind the external photosphere. The same is true for the intermediate-width Hα in post-maximum phases. At early times, however, when the post-shock Hα is hidden behind the photosphere, the emergent Hα line is dominated by a completely different emission mechanism. As we proposed above, the intermediate-width Hα emission at early times probably arises from photoionized pre-shock circumstellar gas and subsequent scattering by thermal electrons. We find, therefore, that red Fe ii lines and the Ca ii triplet are more robust tracers of the direct CSM interaction luminosity than Hα, since intermediate-width Hα can be produced by a different mechanism. The Ca ii lines have no narrow emission at any epoch.

The red Fe ii lines are too blended to derive individual line-profile shapes, but the profile of Ca ii λ8662 is displayed in Figure 23 as seen in all our epochs of LRIS spectra. Like Hα, Ca ii λ8662 has a bright intermediate-width emission component with a half-width of roughly 2,000 km s⁻¹ and broader line wings. Although the strength of the line changes, its width does not change much. Ca iiλ8662 also has a sharp narrow blueshifted absorption component at roughly −200 km s⁻¹, and is missing the blue side of the line, suggesting rather strong broad blueshifted absorption. Based on the similarity of their red emission wings, Hα and Ca ii probably trace similar regions in the dense shell.

Zoom In Zoom Out Reset image size

The presence of strong intermediate-width lines of Ca ii from the post-shock gas is of considerable interest with regard to SN 2008S and the similar transient in NGC 300, both of which showed particularly strong emission of Hα, [Ca ii], and the Ca ii near-IR triplet (Smith et al. 2009b; Bond et al. 2009; Berger et al. 2009). SN 2006gy also may show the [Ca ii] lines, although we cannot reliably separate them from [O ii] and Fe ii lines.

Lastly, we note the presence of broad absorption features at 6700–7000 Å, where the red wing of Hα meets the continuum. Signs of this absorption first appear on day 127, and the absorption strengthens with time on days 145 and 153. It subsides at subsequent epochs and disappears by day 237. The change in strength of the broad absorption features follows the strength of the emission in the red Fe ii and Ca ii lines noted above. Prominent absorption at these wavelengths appears to be a common feature in late-time spectra of luminous SNe IIn such as SNe 1997cy, 1999E, 2002ic, and 2005gj. For example, the day 98 spectrum of SN 2005gj is shown in Figure 5.⁷

The identification of these broad absorption lines is uncertain. Broad lines near the wavelengths of these features are seen in SN photospheres of Types Ia, Ib, and Ic at late times, although the different expansion speeds make a direct comparison with SN 2006gy difficult. This feature was part of the rationale for associating SNe 2002ic and 2005gj with SNe Ia, although Benetti et al. (2006) have questioned this association and suggested SNe Ic as the culprit instead. SN 2006gy informs this debate for the following reason: In the case of SN 2006gy, these absorptions cannot be due to features in an underlying SN photosphere, because they are too strong.

The fact that features are seen in absorption at such late phases requires the presence of some substantial source of background continuum light. At its strongest, the broad absorption reaches a depth of roughly 12%–15% of the continuum level (Figure 22), requiring that the underlying source that is being absorbed has a luminosity of (1–2) × 10⁹ L_☉ at the late phases of 120–170 days when this absorption is seen. No standard SN photosphere is that luminous at such late times, so these lines cannot arise in a simple scenario of normal SN ejecta being seen through the thinning CSM interaction region. The luminosity may come from the reverse shock or radioactive decay, as we discuss further in Section 4.1. More complicated effects may play a role as well, such as "top-lighting" (e.g., Branch et al. 2000), caused by irradiation of the SN ejecta by photons emitted from the CSM interaction region. Since SNe with strong CSM interaction like SN 2006gy can mimic features seen in other SNe, this casts doubt on the association of such features with an underlying SN Ia photosphere in other luminous SNe IIn. This would seem to support claims by Benetti et al. (2006) that this class of objects could be SNe other than Type Ia.

Regardless of the identification, from the broad line profiles and the fact that they do not appear until late times, it is likely that these lines arise from SN ejecta that have just crossed or are just about to cross the reverse shock. As we note below, the speeds of unshocked SN ejecta and the expansion speed of the CDS in the post-shock gas are comparable at these late times.

Altogether, the extraordinary spectral properties of SN 2006gy can be understood in the context of CSM interaction where the radiating layers evolved through three key phases in the first year, as follows.

• 1.  
  Early times (days 0–90). At these phases before and up to the time of maximum luminosity, the photosphere is outside the forward shock and the emergent spectrum is generated in the pre-shock CSM. Radiation diffusing from the shocked shell propagates forward and pre-ionizes the opaque CSM, leading to a smooth continuum plus Balmer lines generated in CSM layers immediately outside that. Consequently, the post-shock CDS and the SN ejecta cannot be seen yet. The intermediate-width Hα line profile arises because narrow lines from the photoionized pre-shock CSM are broadened by multiple scattering with thermal electrons in the opaque gas (e.g., Chugai 2001). This hypothesis is consistent with the observed Hα/Hβ intensity ratio indicative of recombination at early times, and the Lorentzian profiles of Hα line wings. At this phase, therefore, the prime luminosity source is diffusion of shock-deposited energy (Smith & McCray 2007) in a manner similar to that described by Falk & Arnett (1977).
• 2.  
  Decline from Maximum Luminosity (days 90–150). During these phases, the photosphere recedes (in mass) past the forward shock and into the CDS. As the photosphere passes this sharp boundary, the consequent transition in spectral morphology from phase 1 to phase 2 may be sudden. The effective value of R_BB begins to fall, not because the radius is actually decreasing, but because the CDS becomes progressively more transparent (Smith et al. 2008a). As the CDS is exposed, we begin to see strong emission from cooling lines in the post-shock regions, such as multiple Fe ii and Ca ii lines, which turn on suddenly after day 90, in addition to broad Hα. Broad blueshifted absorption features (see Section 4.2 below) appear during this time as the photosphere is now inside the CDS, and outer layers between the CDS and the forward shock can be seen in absorption. Broad emission-line wings and the blueshifted absorption show that the expansion velocity of the CDS does not slow during this time period, indicating self-similar expansion. The broad line-wing shape is probably a composite of the true kinetic line width (about 4000 km s⁻¹ as indicated by the blueshifted absorption) plus a contribution from electron scattering within the clumpy CDS itself (not in dense CSM as in stage 1; see Dessart et al. 2009). The continuum luminosity during this time is supplied by a mix of fading diffusion from the gradually thinning shocked shell, plus ongoing CSM interaction luminosity. The continuum effective temperature remains roughly constant. The speed of the pre-shock CSM appears to be increasing during this time, suggesting a Hubble-like flow in the episodic pre-SN mass loss that occurred ∼8 yr before explosion.
• 3.  
  Drop in Late-Time CSM Interaction (days 150–240). Beginning around day 150, diagnostics of the post-shock cooling luminosity, such as Hα, Fe ii, and Ca ii lines (see Figure 22), show a precipitous and continuing decline. At these late phases the emitting layers must be optically thin and getting thinner with time, so the weakness of these lines cannot be attributed to high optical depth as before. Instead, the decline probably indicates that the CSM interaction luminosity itself is dropping as the forward shock reaches the outer boundary of the dense pre-SN shell ejected 8 yr earlier. A key point is that during the transition from stage 2 to stage 3, the narrow absorption components formed in the pre-shock CSM were observed to weaken and disappear in some cases, signaling lower pre-shock densities upstream, and heralding the inevitable approach of the outer boundary of the dense pre-shock CSM. This confirms that the reason the Hα, Fe ii, and Ca ii line luminosities dropped at late times was because the SN ran out of CSM to sweep up, not because the SN ran out of energy or recombined.

The drop in CSM interaction at days 150–240 explains why the late-time spectrum observed 1 yr later showed no sign of ongoing CSM interaction. Hα was observed to be about 400 times weaker than in SN 1988Z or SN 2006tf at a comparable epoch (Smith et al. 2008b).

Returning to the spectral behavior during stage 3 at days 150–240, then, it remains quite perplexing why the continuum luminosity did not drop by an amount comparable to the drop in Hα, Fe ii, and Ca ii as it did in SN 1994W (Chugai et al. 2004). It should have dropped if it were generated primarily by CSM interaction. Instead, the red-continuum luminosity appears to have slowed its decline during this stage and even seemed to level off at late times. The SN was still quite bright at these late times—it was still as luminous as the peak of a normal Type Ia—so this extra late-time luminosity source after day 200 is nontrivial.

Given the difficulty imposed by the nearby host-galaxy nucleus, it is conceivable that some of the blue-wavelength light after day 200 is contaminated by host-galaxy light, but red wavelengths show a clear detection of the SN. In a companion paper (Miller et al. 2009b), we confirm that the lingering emission detected 1–2 yr later is indeed due to a light echo as proposed by Smith et al. (2008b). However, a reflected-light echo cannot explain the extra luminosity around days 200–240, because (1) the continuum shape observed then was redder than that at the time of peak luminosity; we would expect reflected light to be the same as or bluer than the original spectrum, as it is at later times (Miller et al. 2009b), and (2) a distant light echo cannot give rise to the broad absorption features seen in the spectrum (Section 3.7).

One possibility for the excess emission at days 200–240 is a re-energized reverse shock, powered by increased density or speed of material crossing the reverse shock and re-heating the CDS from the inside. Even if the CSM density drops so that the forward shock no longer encounters a significant obstacle, the massive CDS that has already formed and is now coasting at 4000 km s⁻¹ provides a wall that may decelerate the faster SN ejecta passing the reverse shock. Alternatively, some of the extra luminosity of (1–2) ×10⁹L_☉ needed for the broad absorption features seen on days 120–170 could be supplied, hypothetically, by radioactive ⁵⁶Co decay from ≲2.5 M_☉ of ⁵⁶Ni. Unfortunately, SN 2006gy was aligned too closely with the Sun after day 240 and we were no longer able to observe it until the later data reported by Smith et al. (2008b), when it had already faded considerably. See Miller et al. (2009b) for additional discussion of the late-time data.

Here we take a more detailed look at the origin of the unusual spectrum of SN 2006gy during the second stage described above, days 90–200 when it showed several overlapping complex, peculiar line profiles. While SN 1994W and SN 1995G showed similar spectral lines at earlier epochs, with narrow emission and absorption, they did not exhibit such prominent broad absorption that gives SN 2006gy its distinct spectral appearance.

A cartoon of a hypothetical line profile for a single line is shown in Figure 24. The narrow and broad absorption, components 1 and 3, are common to many lines in the spectrum. The narrow and broad emission, components 2 and 4, are seen in some lines and not in others, depending on the ionization state, temperature, and density of the gas. For example, (1) both emission components are absent in the red Si ii doublet, (2) only the broad emission is present in Fe ii λ5018 and Ca ii λ8622, and (3) both narrow and broad emission are seen in Hα. A series of these line profiles seen in superposition, having different strengths of absorption or emission, can lead to the high-resolution spectrum of SN 2006gy seen in Figure 17. Whether self-consistent radiative transfer calculations can account for the appearance of the spectrum is a more difficult challenge.

Zoom In Zoom Out Reset image size

Figure 24 also indicates the likely places where components 1, 2, 3, and 4 may be produced in a simplified CSM-interaction scenario (by "simplified," we mean that we do not show the clumping, instabilities in the CDS, or large-scale asymmetry that are all possible). Because of their low velocities, the narrow components 1 and 2 must arise in the pre-shock CSM, either in absorption along our line of sight (1) or in emission away from our direct line of sight (2). Such narrow features are common in spectra of SNe IIn with sufficient resolution.

The broad (actually intermediate-width) emission component (4), which is a common characteristic of SNe IIn, is formed in the CDS expanding at ∼4000 km s⁻¹ during this time interval (days 90–200). We showed that the intermediate-width Hα emission component can be approximated as a symmetric composite Gaussian with two components of roughly equal intensity, one with FWHM = 5200 km s⁻¹ and one with FWHM = 1800 km s⁻¹. While these distinct velocities are somewhat artificial, this emission component likely arises as a result of the intrinsic kinematic line profile plus broadening of the line wings due to electron scattering within the optically thick CDS itself. Note that electron scattering within the line-forming region in the CDS, as suggested by Dessart et al. (2009), is different from the electron scattering that broadens the wings of narrow lines formed in the CSM that we discussed earlier (Section 3.3.1; see Chugai 2001). We suspect that both are at work in SN 2006gy at different times—one in the early-time spectra when the photosphere was outside the shock, and one during the decline from peak luminosity when the photosphere and Hα-emitting region were in the CDS. The idea that electron scattering plays a role in forming the broad wings of Hα is supported by the fact that they are unchanging with time, and that the blue Hα emission wing extends to faster speeds (5000–6000 km s⁻¹) than the blue edge of Hα absorption at 4000 km s⁻¹. Different lines, like Fe ii and Ca ii, originate from different optical depths within the CDS and will therefore be modified to differing degrees by electron scattering. Whether different line-profile shapes are consistent with this hypothesis is another challenge for detailed calculations. Dessart et al. (2009) have shown, for example, that even different H i lines have different line wings for this reason.

The broad blueshifted absorption in SN 2006gy (component 3) is unusual among SNe IIn. The dramatic difference in blueshifted Hα wings of SN 2006gy and SN 2006tf was illustrated in Figure 8 of Smith et al. (2008a). While many SNe IIn do show blueshifted absorption, it is usually from narrower components formed in the CSM. The absorption in component 3 of the SN 2006gy line profiles is far too broad to arise in the CSM, and must therefore reside between the FS and the photosphere. The most likely location is in outer layers of the CDS, as long as the continuum photosphere is in deeper layers. The source of the continuum could, in principle, be a combination of light from the underlying SN ejecta or continuum generated in deeper layers of the CDS which absorb X-rays generated in the reverse shock.

Figure 25 summarizes several physical aspects of SN 2006gy interpreted in the format of a standard CSM-interaction model. This is not a hydrodynamic model, but rather a set of parameters derived from observations with a few simplifying assumptions. Two key observed values are the SN's bolometric luminosity and the radius of the CDS which emits the intermediate-width Hα line. Solid circles in Figures 25(a) and (b) are just the bolometric luminosity and R_BB from Figure 7, while solid black curves are simple continuous approximations of the data. We assume that the true shock radius R continues to increase at roughly constant speed even though the apparent R_BB begins to decrease around day 115, as explained in Section 3.2. Gray lines in Figure 25(b) represent trajectories for unperturbed SN ejecta.

Zoom In Zoom Out Reset image size

Figure 25(c) shows three relevant expansion speeds. At each time t, the speed of the fastest freely expanding SN ejecta that reach the reverse shock (assumed to be the same as R) is simply V_ejecta = R/t, shown by the dotted curve. The speed of the CDS, V_CDS, plotted in Figure 25(c) is just the instantaneous value of $\dot{R}$. For our purposes here, we assume the post-shock cooling zone in this radiative shock to be very thin, so that R ≈ R_FS ≈ R_RS. The solid line representing V_CDS in Figure 25(c) is consistent with the relatively constant observed speeds of 4000–5000 km s⁻¹ seen in Hα from the time of peak luminosity onward. Also, these values of V_ejecta and V_CDS are consistent, at least qualitatively, with expectations of hydrodynamic models of SNe IIn (e.g., Chugai & Danziger 2003). The dashed line in Figure 25(c) is the approximate speed of the progenitor's wind or CSM shell, V_CSM, that is being overtaken by the blast wave (V_CSM is multiplied by 10 for display). This is inferred from observations of the blue edge of the narrow absorption component of Hα, shown in Figure 14. One exception is that we have set the minimum speed at small radii to 150 km s⁻¹; this assumption is irrelevant except for the role that V_CSM plays in deriving $\dot{M}$ below.

To calculate the remaining parameters in Figure 25(d), we adopt the standard equation for the maximum luminosity that can be generated by CSM interaction, assuming 100% efficiency, given by

$$\begin{equation} L = \case{1}{2} w V_{\rm CDS}^3, \end{equation} \tag{ 1 }$$

where w is the wind density parameter w = 4πR²ρ, or $w = \dot{M} / V_{\rm CSM}$, and V_CDS is the value for the evolving speed of the CDS derived from $\dot{R}$ above. The efficiency may be substantially less than 100% (a value of 30%–50% is probably more realistic), which would act to raise all the derived values plotted in Figure 25(d). The wind density parameter is plotted here in units of 10¹⁸ g cm⁻¹ (this is quite a large fiducial value, even for luminous SNe IIn). The true wind density is ρ = w/4πR², plotted here in units of 10⁻¹⁴ g cm⁻³, and the mass-loss rate of the progenitor star inferred at a given time is $\dot{M} = w V_{\rm CSM}$. The value of M_Tot given by the dotted line is the cumulative CSM mass that has been swept up by the shock and now resides in the CDS.

A key assumption here is that direct radiative cooling from the post-shock gas supplies the observed instantaneous value of the luminosity. We have argued against this on observational grounds (Paper I), and have suggested that some of the peak luminosity may in fact be due to the delayed escape or diffusion of shock-deposited thermal energy from earlier CSM interaction. If true, the net result is that the values of w, ρ, and $\dot{M}$ in Figure 25(d) will be overestimated around the time of peak luminosity, but severely underestimated before then. The end value of M_Tot, however, is unaffected because Equation (1) is linear in L and w.

We see from Figure 25 that the swept-up mass needed to provide the luminosity of SN 2006gy in a CSM-interaction scenario is already 10 M_☉ (or perhaps 2–3 times more if the efficiency in converting kinetic energy to visual light is not 100%) by the time of peak luminosity, and almost 20 M_☉ by 200 days after explosion. This large mass dictates that the dense shell must have very high optical depths and a long diffusion time. The ∼10 M_☉ swept up by the time of peak luminosity matches the value adopted by Smith & McCray (2007) in such a model. It is therefore likely that the low CSM density before day 50 (i.e., at R ≲ 4 × 10¹⁵ cm) plotted in Figure 25(d) is erroneous, since it stems from the probably false assumption (at early times) that the observed luminosity directly traces CSM interaction. In reality, high optical depths prevent shock-deposited thermal energy from escaping promptly, leading to a value from Equation (1) that is too low.

The need for higher densities at inner radii (within 4 × 10¹⁵ cm) is evident from a separate line of reasoning as well. The spectrum at early times (e.g., day 36) arises because the photosphere is outside the forward shock, since shock luminosity must diffuse out through the opaque CSM shell. In the process, multiple scattering with thermal electrons broadens the narrow Hα emission from the photoionized pre-shock gas, causing symmetric profiles. This picture, however, necessitates pre-shock CSM density on day 36 exceeding that on days 90–100 when the pre-shock CSM is no longer opaque. On day 36 we found an optical depth for the pre-shock CSM of roughly τ ≈ 15, whereas on days 90–100 it must have been ≲1 because the CDS is seen. The inferred CSM densities for days 36 and 90 in Figure 25(d), however, are comparable, which cannot be the case. This discrepancy can be reconciled if we relax the assumption that the observed luminosity traces the level of CSM interaction, and is instead due to delayed diffusion at early times. Much higher CSM densities, probably above 10⁻¹³ g cm⁻³, are required at inner radii of 2 × 10¹⁵ cm. Indeed, CSM densities of 10⁻¹³ to 10⁻¹² g cm⁻³ were required at these inner radii in models of SN 1994W by Chugai et al. (2004), which showed qualitatively similar spectral properties although at lower luminosity. SN 2006gy probably requires even higher densities at these inner radii.

In any case, by late times (200–250 days after explosion,) the total CSM mass that has been swept up into the CDS by the shock is 18–19 M_☉. This agrees with the value predicted in the models of Woosley et al. (2007) for a pulsational pair-instability ejection. With a final coasting velocity of ∼4000 km s⁻¹, the kinetic energy remaining in the CDS when the strongest CSM-interaction phase has ended is then at least (2.9–3.0) × 10⁵¹ erg. Combined with the integrated radiated energy of E_rad = (2.3–2.5) ×10⁵¹ erg (Section 3.2), the initial kinetic energy of SN 2006gy was more than 5 × 10⁵¹ erg and the efficiency in converting energy into visual light was <50%. This efficiency, in turn, suggests SN ejecta whose mass was comparable to that of swept-up CSM, implicating a very massive progenitor star.

Qualitatively, the CSM-interaction scenario we propose for SN 2006gy is actually quite analogous to scaled-up models for SN 1994W and SN 1995G by Chugai et al. (2004) and Chugai & Danziger (2003), respectively. In both cases, the authors claimed that the observed luminosity was due to a combination of internal energy leakage from an extended pre-SN shell, plus subsequent luminosity from CSM interaction. This is qualitatively identical to our proposed model for SN 2006gy, where the prime luminosity source at early times is the slow diffusion of shock-deposited energy in an unbound opaque CSM envelope ejected ∼8 yr before the SN (see Smith & McCray 2007). In the case of SN 1994W the pre-SN shell was ejected 1.5 yr before explosion in a 2 × 10⁴⁸ erg event (Chugai et al. 2004), and for SN 1995G it was ejected 8 yr before in a 6 × 10⁴⁸ erg event (Chugai & Danziger 2003). This is reminiscent of the SN 2006gy pre-SN outburst, which occurred ∼8 yr before with ≳10⁴⁹ erg. Both SN 1994W and SN 2006gy showed evidence that the strength of CSM interaction plummeted as the blast wave overtook the outer boundary of this CSM shell.

The primary difference between SN 2006gy and the other two events is that, quantitatively, SN 2006gy is a much more extreme case, with 20–50 times more mass ejected in its precursor event, and a more energetic SN explosion. Consequently, diffusion through the opaque CSM shell made a more important contribution to the total luminosity as SN 2006gy remained fully optically thick (i.e., R_BB was equal to the true shock radius) out to much larger distances. This is why SN 2006gy had a smaller Hα equivalent width (see Figures 5 and 12). Qualitatively, all three SNe showed similar absorption features in their spectra (Figure 5). As discussed above, some other SNe IIn like SN 1998S and SN 2005gl passed through similar phases that lasted only a short time, implying that they are very slimmed-down versions of similar phenomena.

We have not invoked a high degree of clumping in the CSM of SN 2006gy, because clumping allows the blast wave to propagate relatively unimpeded through the lower-density interclump regions (i.e., the shock remains fast), it facilitates the escape of copious X-rays, and it allows an observer to see radiation from the underlying SN photosphere. These traits are not observed in SN 2006gy.⁸ This is a key difference between SN 2006gy and models for less luminous (but X-ray bright) SNe IIn like SN 1988Z and SN 2005ip, where clumpy CSM was invoked for these reasons (Chugai & Danziger 1994; Smith et al. 2009a). This is also a difference between our model for SN 2006gy and that of Agnoletto et al. (2009), who assumed strong clumping.

In Paper I we noted that the extraordinary peak luminosity of SN 2006gy requires exceptional physical conditions unlike those of previous SNe. If radioactivity is the luminosity source, then an unusually large mass of ⁵⁶Ni near 10 M_☉ would be required, which in turn would necessitate a PISN (e.g., Barkat et al. 1967; Rakavy & Shaviv 1967; Bond et al. 1984; Heger & Woosley 2002). If, on the other hand, CSM interaction were the engine, then a surprisingly large mass of CSM would be needed, comparable to the 10–20 M_☉ of H-rich material ejected in giant LBV eruptions such as the 1843 event of η Carinae (Paper I; Smith et al. 2003) or some event like theoretical pulsational pair-instability eruptions (Woosley et al. 2007). In Paper I we were unable to choose between those two possibilities.

After considering the spectral evolution of SN 2006gy in detail, we find that most of its physical and morphological properties can be interpreted in the context a two-component CSM-interaction model. The apparent contradictions of a simple CSM-interaction model that were noted in Paper I can be reconciled with observations by pushing the physical conditions—in particular the optical depth and CSM mass—to extremes. High optical depths lead to a situation where the luminosity of SN 2006gy is powered by a two-component engine: radiative diffusion from ∼10⁵¹ erg of shock-deposited thermal energy dominates at early times, while an increasing fraction at later times is supplied by the normal scenario of direct radiative post-shock cooling. This two-component CSM interaction engine alleviates the need for an internal energy source such as radioactive decay from 10 M_☉ of ⁵⁶Ni because it reconciles the high CSM interaction luminosity with the seemingly paradoxical weakness of CSM interaction diagnostics like Hα and X-ray emission.

Two final points are worth emphasizing, however. First, the extremely dense and massive CSM required for SN 2006gy does seem to be quite well explained by the predictions of Woosley et al. (2007) for a pulsational pair-instability ejection. This event has the same underlying instability, but results in a nonterminal precursor explosion. It is expected to occur for stars with initial masses in the range 95–135 M_☉, just shy of the extremely high masses required for a true PISN (see Heger et al. 2003). Second, the late-time (day 200–240) luminosity of SN 2006gy is still a problem, and it remains difficult to rule out some contribution from an internal energy source. The late-time luminosity is sustained even though diagnostics of the CSM interaction (Hα, Fe ii, Ca ii) plummet. This luminosity at 200–240 days cannot be due primarily to a light echo (which does dominate 1–2 yr later; Smith et al. 2008b; Miller et al. 2009b).

The high peak luminosity of SN 2006gy has been exceeded by two recent SNe, SN 2005ap (Quimby et al. 2007) and SN 2008es (Miller et al. 2009a; Gezari et al. 2009). The interesting twist, though, is that neither of these had a Type IIn spectrum, unlike all the luminous runners-up. It is therefore not obvious that their luminosity is generated by CSM interaction.

In this paper we have found that SN 2006gy exploded in a dense two-component CSM, with a massive opaque inner shell within a radius of (2–4) ×10¹⁵ cm, plus a more extended circumstellar envelope with an outer boundary of roughly 10¹⁶ cm. We have concluded that it was largely the diffusion of radiation from interaction with the inner opaque shell that powered its rise to peak luminosity, as proposed by Smith & McCray (2007), while the continued CSM interaction with the more extended shell powered some of the continuum luminosity at late times and determined the Hα luminosity, soft X-ray luminosity, and unusual absorption and emission-line profiles. Photoionization of this outer shell produced the narrow emission-line components in the spectrum. If this outer shell had not been present, however, SN 2006gy would not have had a Type IIn spectrum.

The outer radius of the outer shell of SN 2006gy at ∼10¹⁶ cm was determined primarily by the maximum velocity of the precursor mass ejection and the fact that it occurred 8 yr before explosion. If that ejection episode had less mass and energy, or if it had occurred a shorter time before the final SN, the resulting CSM envelope may have been less massive and less extended, yielding a luminous SN II powered mainly by diffusion. Similarly, with CSM that is more extended and more optically thin, one can achieve a long-lasting SN IIn like SN 1988Z. By varying the physical properties of the precursor ejection—and hence, changing the relative contribution of the optically thick diffusion and the ongoing optically thin CSM interaction engines—we can understand the diversity of observed properties in the most luminous SNe. A lower CSM mass would have resulted in faster final expansion speeds due to conservation of momentum. The lower CSM mass would also cause a faster diffusion time—and hence, more rapid rise and decay times—for interaction with the opaque shell. Without the more extended shell, there would have been no ongoing optically thin CSM interaction to drive the Type IIn spectrum. Thus, with a modified set of pre-SN conditions, one can see how a similar CSM interaction could power both SN 2005ap and SN 2008es. We therefore consider it unlikely that these events were true PISNe. Observed properties, however, are consistent with pre-SN shell ejections triggered by a pulsational pair instability, or perhaps some other as-yet unidentified explosive instability.

In Paper I and in this work, we have noted a number of similarities between the inferred physical properties of the mass ejection that occurred 8 yr before SN 2006gy and the observed properties of giant LBV eruptions like the 1843 event of η Carinae. The most important of these are the total ejected mass of 10–20 M_☉ (Smith et al. 2003), launched at speeds of 100–600 km s⁻¹ with a Hubble-like velocity law indicating a brief event (Smith 2006a), and the H-rich composition. We have also noted that these properties, and the fact that the outburst occurred within a decade of the SN, seem equally consistent with a precursor event caused by the pulsational pair instability (Woosley et al. 2007).

We wish to emphasize that these two scenarios are not necessarily mutually exclusive. One is observational (the comparison to observed LBV eruptions like η Car), and the other is a theoretical idea (the pulsational pair instability). We cannot rule out the possibility that they are, in fact, the same phenomenon, since we still do not know what triggered the 1843 outburst of η Car. After all, most of the observed properties of η Car's giant eruption are consistent with expectations of the pulsational pair instability, except that η Car has not yet met its final demise even though it suffered similar previous outbursts ∼1000 yr ago. Judging by the additional massive ∼10–20 M_☉ dust shell required for the late-time IR light echo of SN 2006gy (Smith et al. 2008b; Miller et al. 2009b), it also appears to have suffered a previous shell ejection event ∼1500 yr ago. Heger & Woosley (2002) have noted that in some cases, the time between successive pair pulsations can reach as much as ∼1000 yr. Furthermore, recent observations of η Car have shown that it even had an explosive component to the event that created a fast (∼5000 km s⁻¹) blast wave out ahead of the most massive material (Smith 2008).

Regardless of whether SN 2006gy was a PISN or instead a sequence of pre-SN shell ejections caused by pair pulsations or some other mechanism, we argued in Paper I that the extreme energy and mass budgets make it difficult to avoid the requirement that SN 2006gy marked the death of a very massive star. The work described here reinforces this view: we find that the amount of CSM required is ∼20 M_☉, and that the mass of SN ejecta must be comparable in order to yield the final efficiency of about 50% in converting kinetic energy to light. Furthermore, our previous study of the late-time IR emission (Smith et al. 2008b) suggested a dust shell with a total gas mass of at least 10 M_☉ ejected 1000–2000 yr ago; this has been confirmed by continued study (Miller et al. 2009b). This mass budget requires that within ∼10³ yr of its final explosion, the progenitor star still retained no less than 50–60 M_☉.

A key point to recall is that SN 2006gy occurred in a galaxy of roughly solar metallicity, so the progenitor star would have shed significant mass during its lifetime and its initial mass was probably above 100 M_☉. This is consistent with models for SN 2006gy as a pulsational pair event (Woosley et al. 2007), which involved a star with an initial mass of 110 M_☉. (Similar requirements appear to be necessary for SN 2006tf, but the mass budgets are more relaxed for SNe 2005ap and 2008es.)

Therefore, it seems impossible that SN 2006gy was a thermonuclear (Type Ia) SN that exploded in a dense H-rich CSM to produce a Type IIn spectrum, as had been suggested for SNe 2002ic and 2005gj (Hamuy et al. 2003; Chugai & Yungelson 2004; Aldering et al. 2006; Prieto et al. 2007), and possibly also SNe 1997cy and 1999E (Germany et al. 2000; Turatto et al. 2000; Rigon et al. 2003). These had been the most luminous SNe known until SN 2006gy. We have noted that in several observed respects, SN 2006gy can be regarded as a more extreme case of this class of objects. It therefore seems possible that not all of them are SNe Ia in dense CSM environments either, supporting the view of Benetti et al. (2006). Trundle et al. (2008) have also expressed doubt that SN 2005gj was a Type Ia event.

If SN 2006gy really was the death of a very massive star, it challenges the standard paradigm that all massive stars at near-solar metallicity should shed their H envelopes and explode as Wolf-Rayet (WR) stars (Conti 1976; Chiosi & Maeder 1986), yielding SNe of Type Ib or Ic and possibly gamma-ray bursts (e.g., Langer et al. 1994; Heger et al. 2003; Woosley & Heger 2006). Instead, it requires that some extremely massive stars retain substantial H envelopes up until the time of core collapse.

This brings to mind the recent recognition that steady line-driven mass-loss rates of O-type stars and WR stars are lower than those included in most models. The winds are clumped, which lowers the mass-loss rate one derives from observations (see Bouret et al. 2005; Fullerton et al. 2006). Smith & Owocki (2006) pointed out that LBV eruptions may help make up some of the deficit, but if insufficient, the star can explode as an LBV with part of its H envelope still intact.

We have argued here and elsewhere (Smith et al. 2007, 2008a, 2008b; Smith 2006b) that the extreme progenitor mass-loss rates required to produce the dense CSM around luminous SNe IIn suggests a link to LBV-like instability in the progenitor stars. LBVs in eruption are the only stars that are observed to have mass-loss rates of the correct order to explain more luminous SNe IIn. Outer shells around at least some SNe, such as the dusty IR echo shell around SN 2006gy (Smith et al. 2008b), imply that these eruption episodes repeat on a timescale of ∼10³ yr, like LBVs. (We have also noted, however, that some of the less luminous SNe IIn, such as SN 2005ip, can potentially be explained with the most extreme cases of RSG mass loss, so not all SNe IIn necessarily come from LBVs; see Smith et al. 2009a, 2009c.)

The possibility of a LBV/SN IIn connection, inferred indirectly from pre-SN CSM properties, has been confirmed by the only case so far where a SN IIn progenitor star has been identified directly in pre-explosion images. In SN 2005gl it was found that the progenitor did in fact resemble a very luminous LBV-like star with high luminosity (Gal-Yam et al. 2007, Gal-Yam & Leonard 2009). Gal-Yam & Leonard found that the progenitor had a temperature consistent with LBVs (i.e., a blue to yellow supergiant, not a RSG) and a very high luminosity of ∼10⁶ L_☉. They also demonstrated that this star vanished after SN 2005gl faded. A massive progenitor with initial mass above ∼50 M_☉ seems to be required. To explain the observed level of CSM interaction in SN 2005gl, Gal-Yam & Leonard (2009) find that the precursor shell ejection event had a mass-loss rate of ∼0.03 M_☉ yr⁻¹, similar to the 1600 AD giant eruption of the famous LBV star P Cygni (Smith & Hartigan 2006). The fact that SN 2005gl itself was not a particularly high-luminosity event compared to other SNe IIn underscores the truly astonishing set of conditions that led to SN 2006gy. The progenitor of SN 2006gy must have been an extreme case of an already extraordinary class of stars.

There are several other circumstantial indications that LBVs may be related to SNe as well, mainly from inferred properties of the H-rich pre-shock CSM (Salamanca et al. 2002; Kotak & Vink 2006; Smith 2006b, 2007; Smith et al. 2007, 2008a, 2008b; Trundle et al. 2008). One argument made by these authors is that the speed and possible variability of the pre-shock CSM is close to the outflow speeds of a few hundred km s⁻¹ observed in LBVs, while being much faster than RSG winds and slower than WR winds. This is true for SN 2005gl as well (see the Appendix). Wind speeds alone are not conclusive, since radiation from the SN itself may accelerate pre-shock CSM to these speeds. This could, in principle, cause the two-component absorption features in SN 2005gj that Trundle et al. (2008) showed to be very similar to Hα profiles in LBVs. While not conclusive on their own, the pre-shock speeds are certainly consistent with the LBV interpretation. Taken together with the large CSM masses needed to power the luminosity (discussed in this work) and the similarity between LBV nebulae and the pre-SN nebula around SN 1987A (Smith 2007), they support the view that in some cases LBV episodes may indeed precede SNe.

If other SNe IIn are also very massive stars that retain their H envelopes until just before death, it suggests a picture where these stars enter a phase of instability shortly before explosion that triggers the ejection of much or all of their remaining H envelopes. The underlying SN explosion, which, hypothetically, may otherwise have been a Type Ib/c, IIb, II-P, or II-L event, expands into that recently ejected massive H envelope to produce a SN IIn. (In this regard, it may be more appropriate to consider Type IIn's as a phenomenon rather than a class of explosions.) One might expect a very massive star nearing core collapse to be unstable if it has retained part of its H envelope, because its outer H envelope would be dangerously close to or may exceed the classical Eddington limit. The attempt to understand this phenomenon is likely to benefit from continued study of the physics of giant LBV eruptions and theoretical investigations of super-Eddington winds (e.g., Owocki et al. 2004; van Marle et al. 2009a), even though it is not yet clear if the two phenomena have the same underlying trigger. Such pre-SN instability might be exacerbated by additional luminosity from final burning stages. A star could therefore be subject to periodic and sudden bursts of mass loss as it tries to shed its massive H envelope. These clues suggest that the final years of very massive stars include sporadic events that essentially determine the end fate of the star. Significant revision is therefore needed to correctly model and predict the fates of the most massive stars beyond our speculation noted here, including dynamical events like LBV eruptions, pulsational pair ejections, or other instabilities associated with late phases.