A PANCHROMATIC VIEW OF THE RESTLESS SN 2009ip REVEALS THE EXPLOSIVE EJECTION OF A MASSIVE STAR ENVELOPE

We initiated our Swift-UVOT (Roming et al. 2005) photometric campaign on 2009 September 10 and followed the evolution of SN 2009ip in the six UVOT filters up until 2013 April. Swift-UVOT observations span the wavelength range λ_c = 1928 Å (w2 filter) to λ_c = 5468 Å (v filter, central wavelength listed; see Poole et al. 2008 for details). Our final photometry is reported in Table 4 and shown in Figure 2.

Zoom In Zoom Out Reset image size

Motivated by the bright UV emission and very blue colors of SN 2009ip we initiated extensive UV spectral monitoring on 2012 September 27, ∼6 days before maximum UV light (2012b explosion). Our campaign includes a total of 22 UVOT UV-grism low-resolution spectra (Figure 3) and two epochs of HST observations (PI: R. Kirshner) shown in Figure 4, covering the period −6 days < t − t_pk < +34 days. Figure 5 shows the UV portion of the spectra, renormalized using the blackbody fits derived in Section 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Observations in the v, b, and u filters were obtained with Swift-UVOT as explained in Section 2.1 (see Table 4). We complement our data set with R- and I-band photometry obtained by amateur astronomers. A single, late-time (t_pk + 190 days) V-band observation was obtained on 2013 April 11.40. We measure V = 19.65 ± 0.02 mag.

We obtained 28 epochs of optical spectroscopy of SN 2009ip covering the time period 2012 August 26 to 2013 April 11 using a number of facilities (see Table 2 in Appendix D). The sequence of optical spectra is shown in Figure 6.

Zoom In Zoom Out Reset image size

We obtained ZYJHK data using the Wide-field Infrared Camera on the United Kingdom Infrared Telescope (UKIRT) from September 23 (t_pk − 10 days) until 2012 December 31 (t_pk + 89 days). Our photometry is reported in Table 6. Additional NIR photometry was obtained with PAIRITEL at the Fred Lawrence Whipple Observatory (FLWO) on Mount Hopkins, Arizona. Figure 2 presents the complete SN 2009ip NIR data set.

In addition to the X-shooter spectra, we monitored the spectral evolution of SN 2009ip acquiring a total of nine epochs of NIR spectroscopy between 2012 September 27 (t_pk − 6 days) and 2012 December 3 (t_pk + 61 days). The complete sequence of NIR spectra is shown in Figure 7. The observing log can be found in Table 3.

Zoom In Zoom Out Reset image size

We observed SN 2009ip at millimeter (∼85 GHz) and radio wavelengths (∼21 and ∼9 GHz) using the Combined Array for Research in Millimeter Astronomy (CARMA; Bock et al. 2006) and the Karl G. Jansky Very Large Array (VLA; Perley et al. 2011). Our monitoring samples the evolution of SN 2009ip between 2012 September 26 (t_pk − 7 days) and 2013 March 9 (t_pk + 157 days). A summary of the observations is presented in Table 7. No source is detected at the position of SN 2009ip during the first 50 days since the onset of the major outburst in 2012 September, enabling deep limits on the radio emission around optical maximum. A source is detected at 8.85 GHz on 2012 November 13 (t_pk + 41 days), indicating a rebrightening of SN 2009ip radio emission at the level of F_ν ∼ 70 μJy. The source position is α = 22:23:08.29 ± 001 and δ = −28:56:52.4 ± 01, consistent with the position determined from HST data. We merged the two observations that yielded a detection to improve the signal to noise and constrain the spectrum. Splitting the data into two 1 GHz slices centered at 8.43 GHz and 9.43 GHz, we find integrated flux densities of F_ν = 60.0 ± 16.7 μJy (8.43 GHz) and F_ν = 100.6 ± 18.9 μJy (9.43 GHz), suggesting an optically thick spectrum. The upper limit of F_ν < 70.5 μJy at 21.25 GHz on 2012 December 2 indicates that the observed spectral peak frequency ν_pk is between 9.43 GHz and 21.25 GHz. A late-time observation obtained on March 9 shows that the radio source faded to F_ν < 9.6 μJy at 9 GHz, pointing to a direct association with SN 2009ip (see Figure 8).

Zoom In Zoom Out Reset image size

We observed SN 2009ip with the Swift-X-Ray Telescope (XRT; Burrows et al. 2005) from 2012 September 4 until 2013 January 1, for a total exposure of 260 ks, covering the time period of −29 days < t − t_pk < +90 days. No X-ray source is detected at the position of SN 2009ip during the decay of the 2012a outburst (t < −11 days) and during the rise time of the 2012b explosion (−11 days < t − t_pk < −2 days). X-ray emission is detected at a position consistent with SN 2009ip starting from t_pk − 2 days, when the 2012b explosion approached its peak luminosity in the UV/optical bands (Figures 8 and 9). Starting from t_pk + 17 days, the source was too faint for XRT. We therefore activated our XMM-Newton program (PI: P. Chandra). XMM observations obtained on 2012 November 3 (t_pk + 31 days) yielded a detection. From January until 2013 April the source was Sun constrained. 10 ks of Swift-XRT data obtained on 2013 April 4.5 (t_pk + 183 days, when SN 2009ip became observable again) showed no detectable X-ray emission at the position of the transient.

Zoom In Zoom Out Reset image size

We use the XMM observations and the Swift-XRT observations obtained around the peak to constrain the spectral parameters of the source and flux calibrate the count-rate light-curve. Both spectra are consistent with a thermal bremsstrahlung model with kT = 60 keV and intrinsic neutral hydrogen absorption ${\rm NH_{{\rm int}}}=0.10^{+0.06}_{-0.05}\times 10^{22}\ {\rm cm^{-2}}$. The spectra are displayed in Figure 10 and show some evidence for an excess of emission around ∼7–8 keV (rest-frame), which might be linked to the presence of Ni or Fe emission lines (see, e.g., SN2006jd and SN2010jl; Chandra et al. 2012a, 2012b).

Zoom In Zoom Out Reset image size

We note that at the resolution of XMM and Swift-XRT we cannot exclude the presence of contaminating X-ray sources at a distance ≲ 10''. Observations obtained with the Chandra X-ray Observatory (PI: D. Pooley) on t_pk + 19 days reveal the presence of a contaminating X-ray source lying ≈6'' from SN 2009ip and brighter than SN 2009ip at that time. SN 2009ip is also detected (D. Pooley 2012, private communication). Our contemporaneous Swift-XRT observations constrain the luminosity of the contaminating source to be ≲ 1.5 × 10³⁹ erg s⁻¹, ≲ 30% the X-ray luminosity of SN 2009ip at peak. We conclude that the contaminating source is not dominating the X-ray emission of SN 2009ip around peak, if stable. The temporal coincidence of the peaks of the X-ray and optical emission of SN 2009ip suggests that the detected X-ray emission is physically associated with SN 2009ip. Given the uncertain contamination, in the following section we conservatively assume L_x ≲ 2.5 × 10³⁹ erg s⁻¹ for the peak X-ray luminosity of SN 2009ip.

Stellar explosions embedded in an optically thick medium have been shown to produce a collisionless shock when the shock breaks out from the progenitor environment, generating photons with a typical energy ≳ 60 keV (Murase et al. 2011; Katz et al. 2011). We constrain the hard X-ray emission from SN 2009ip using Swift Burst Alert Telescope (BAT; Barthelmy et al. 2005; 15–150 keV energy range) observations obtained between 2012 September 4 (t_pk − 29 days) and 2013 January 1 (t_pk + 90 days). Analyzing the data acquired around the optical peak we find evidence for a marginal detection at the level of 3.5σ in the time interval −0.8 days < t − t_pk < +0.2 days (corresponding to 2012 October 2.2–3.2). The simultaneity of the hard X-ray emission with the optical peak is intriguing. However, given the limited significance of the detection and the known presence of a non-Gaussian tail in the BAT noise fluctuations (H. Krimm 2012, private communication), we conservatively use F < 7 × 10⁻¹⁰ erg s⁻¹ cm⁻² (L < 8 × 10⁴⁰ erg s⁻¹) as the 5σ upper limit to the hard X-ray emission from SN 2009ip around maximum light.

GeV photons are expected to arise when the SN shock collides with a dense circumstellar shell of material, almost simultaneous with the optical light-curve peak (Murase et al. 2011; Katz et al. 2011). We searched for high-energy γ-ray emission from SN 2009ip using all-sky survey data from the Fermi Large Area Telescope (LAT; Atwood et al. 2009), starting from 2012 September 3 (t_pk − 30 days) until 2012 October 31 (t_pk + 28 days). We find no significant emission at the position of SN 2009ip. Assuming a simple power-law spectrum with photon index Γ = 2, the typical flux upper limits in one-day intervals are ≲ (1–3) × 10⁻¹⁰ erg cm⁻² s⁻¹ (100 MeV–10 GeV energy range, 95% c.l.). Integrating around the time of the optical peak (−2 days < t − t_pk < +4 days) we find F < [2.1, 1.9, 3.6] × 10⁻¹¹ erg cm⁻² s⁻¹ for three energy bands (100 MeV–464 MeV, 464 MeV–2.1 GeV, and 2.1 GeV–10 GeV).

The Hα line profile experienced a dramatic change in morphology. Figure 14 shows the Hα line at representative epochs: at any epoch the Hα line has a complex profile resulting from the combination of a narrow component (FWHM < 1000 km s⁻¹), intermediate/broad width components (FWHM > 1000 km s⁻¹) and blue absorption features with clearly distinguished velocity components. Emission and absorption components with similar velocity are also found in the Hβ and Hγ line profiles (Figure 15). The evolution of the line profile results from changes in the relative strengths of the different components in addition to the appearance (or disappearance) of high-velocity blue absorption edges (Figure 16). The most relevant spectral changes are detailed below.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

By t_pk − 7 days the broad components dominating the line profile 10 days before (Mauerhan et al. 2013) have weakened to the level that most of the emission originates from a much narrower component with FWHM ≈ 1000 km s⁻¹. Absorption from high-velocity material (v ≈ −5000 km s⁻¹, measured at the minimum of the absorption feature) is still detected when the 2012b explosion luminosity is still rising. The high-resolution spectra collected on t_pk − 5 and t_pk − 4 days allow us to resolve different blue absorption components with central velocities v ≈ −2200 km s⁻¹, ≈ − 4000 km s⁻¹, ≈ − 5300 km s⁻¹, and ≈ − 7500 km s⁻¹ with σ ≈ 300–500 km s⁻¹. These absorption features are detected in the Hβ and Hγ lines as well (Figure 15). The width of the narrow component of emission decreases to FWHM ≈ 280 km s⁻¹.

On 2012 September 30 (t_pk − 3 days) SN 2009ip approaches its maximum luminosity (Figure 11). The Hα line is well modeled by the combination of two Lorentzian profiles with FWHM ≈ 240 km s⁻¹ and FWHM ≈ 2600 km s⁻¹. We find no clear evidence for absorption components. Interpreting the broad wings as a result of multiple Thomson scattering in the circumstellar shell of the narrow-line radiation (e.g., Chugai 2001; Dessart et al. 2009; Humphreys et al. 2012) suggests that the optical depth of the unaccelerated circumstellar shell envelope to Thomson scattering is τ ∼ 3. A Thomson-scattering optical depth τ ∼ 2–4 is also expected if the continuum originates at locations where the thermalization depth is ≈1 (Davidson 1987, ³⁷).

High-velocity absorption features in the blue wing of the Hα line progressively reappear as the luminosity enters its declining phase. The Hα line exhibits a combination of narrow (FWHM ≈ 340 km s⁻¹) and broad (FWHM ≈ 2000–3000 km s⁻¹) components, with the broad component becoming more prominent with time. At t_pk + 11 days (time of the third bump of the bolometric light-curve, Figure 11), high-velocity absorption features reappear, with absorption minima at v ≈ −12, 000 km s⁻¹ and v ≈ −8000 km s⁻¹ (σ ∼ 1000 km s⁻¹). The low-velocity P Cygni absorption is also detected at v ≈ −1200 km s⁻¹.

A spectrum obtained on t_pk + 18 days shows the development of an even stronger broad emission component with FWHM ≈ 9400 km s⁻¹. We also find clear evidence for a deep minimum at v ≈ −10, 000 km s⁻¹ with edges extending to v ≈ −14, 000–15, 000 km s⁻¹. The broad emission component keeps growing with time and dominates the line profile at t_pk + 42 days. By t_pk + 51 days, the broad component reaches FWHM ≈ 13, 600 km s⁻¹. High-velocity absorption features are still detected at v ≈ −12, 500 km s⁻¹, v ≈ −10, 000 km s⁻¹, and v ≈ −7000 km s⁻¹.

At t_pk + 79 days we find a less prominent broad component with FWHM ≈ 8200 km s⁻¹. A spectrum obtained at t_pk + 101 days confirms this trend (broad component FWHM ≈ 7500 km s⁻¹): the bulk of the absorption is now at lower velocities v ≈ −3100 km s⁻¹. At t_pk + 190 days the blue-shifted absorption is found peaking at even lower velocities of v ≲ −2400 km s⁻¹, and the "broad" (now intermediate) component has FWHM of only ≈2000 km s⁻¹.

Finally, comparing the H Paschen and Brackett emission lines using our two highest resolution spectra collected around the peak (narrow-line emission dominated spectrum at t_pk − 3 days) and 28 days after peak (when broad components start to emerge, see Figures 19 and 20), we find that for both epochs the line profiles are dominated by the narrow component (FWHM ≈ 170 km s⁻¹) with limited evolution between the two.

We conclude by noting that, observationally, t_pk + 17 days (i.e., 2012 October 20) marks an important transition in the evolution of SN 2009ip: around this time the broad Hα component evolves from FWHM ∼ 3000 km s⁻¹ to FWHM ∼ 10, 000 km s⁻¹ (Figure 16); the photospheric radius R_HOT flattens to R_HOT ∼ 1.6 × 10¹⁵ cm while the temperature transitions to a milder decay in time (Section 3, Figure 11).

He i lines are not unambiguously detected on 2012 August 26. They are however detected one month later, ∼3 days after SN 2009ip rebrightened. As for the H Balmer lines, high-resolution spectroscopy obtained at t_pk − 5 and t_pk − 4 days shows the appearance of multiple absorption components on the blue wing of the He i λ5876 and λ7065 lines, with velocities v ≈ −2000 km s⁻¹, ≈ − 4800 km s⁻¹, and ≈ − 7000 km s⁻¹ (to be compared with Figure 15). High-velocity absorption features disappear by t_pk − 3 days, when the He i lines show the combination of a narrow plus intermediate profiles with FWHM ≈ 240 km s⁻¹ and FWHM ≈ 2000 km s⁻¹, respectively.

Starting from t_pk − 3 days He i features become weaker until He i λ7065 is not detected at t_pk + 28 days (Figure 19). He i later reappears at t > t_pk + 43 days showing the broad/intermediate component only (FWHM ≈ 2500 km s⁻¹ at t_pk + 63 days). At t_pk + 101 days He i λ7065 shows an intermediate-broad emission profile with FWHM ≈ 3000 km s⁻¹. Roughly 100 days later He i 7065 Å is clearly detected with considerably narrower emission (FWHM ≈ 1000 km s⁻¹). He i λ6678 also reemerges on the red wing of the Hα profile (Figure 30).

A number of Fe ii lines have been observed during previous SN 2009ip outbursts (both in 2009, 2011, and the 2012a outburst; see Pastorello et al. 2013, their Figures 5 and 6). The Fe responsible for this emission therefore preexistents the 2012 explosion. Fe ii is not detected during the 2012b outburst until t_pk + 17 days (Figure 17). At t_pk + 28 days the Fe ii lines λ5018 and λ5169 (multiplet 42) have FWHM ≈ 240 km s⁻¹. As a comparison, the FWHM of the narrow Hα component from the same spectrum is ≈170 km s⁻¹. By t_pk + 63 days the Fe ii emission lines develop a P Cygni profile (Figure 17), with v ≈ −1000 km s⁻¹, possibly extending to v ≈ −4000 km s⁻¹.

Zoom In Zoom Out Reset image size

Starting from ∼30 days after peak, NIR emission from the Ca ii triplet λλ8498, 8542, 8662 progressively emerges (Figure 18; see also Fraser et al. 2013, their Figure 4). The appearance of this feature is typically observed during the evolution of Type II SN explosions (see, e.g., Pastorello et al. 2006). No previous outburst of SN 2009ip showed this feature (2012a outburst included; see Pastorello et al. 2013); no broad Ca ii triplet feature has ever been observed in an LBV-like eruption, either. Figure 18 also shows the emergence of broad absorption dips around 8400 Å and 8600 Å, which likely results from the combination of O i and Ca ii.

Zoom In Zoom Out Reset image size

High-velocity, broad absorption features appear in our spectra starting nine days after peak (see yellow bands in Figures 7, 17, 19, and 20). Absorption features of similar strength and velocity have never been associated with an LBV-like eruption to date and are more typical of SNe (Figure 21). These absorption features are unique to the 2012b explosion and have not been observed during the previous outbursts of SN 2009ip (see Smith et al. 2010b; Foley et al. 2011; Pastorello et al. 2013).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

As the photosphere recedes into the ejecta it illuminates material moving toward the observer with different velocities. We identify He i, Na i D and H i absorbing at three typical velocities (Figure 22). The blue absorption edge of He i plus Na i D extends to v ≈ 18, 000 km s⁻¹, as noted by Mauerhan et al. (2013). High-velocity v ∼ −12, 000 km s⁻¹ absorption appears first, around t_pk + 9 days, followed by the v ∼ −5500 km s⁻¹ absorption around t_pk + 28 days, which in turn is followed by slower material with v ∼ −2500 km s⁻¹, seen in absorption starting only from ∼t_pk + 60 days. This happens because material with lower velocity naturally overtakes the photosphere at later times. Material moving at three distinct velocities provides an argument against a continuous distribution in velocity of the ejecta and suggests instead the presence of distinct shells of ejecta expanding with typical velocity v ∼ −12, 000 km s⁻¹, v ∼ −5500 km s⁻¹, and v ∼ −2500 km s⁻¹.

Zoom In Zoom Out Reset image size

We do not find evidence for strong spectral evolution at UV wavelengths (Figures 3 and 5). As time proceeds the Fe iii absorption features become weaker while Fe ii develops stronger absorption features, consistent with the decrease of the photosphere temperature with time (Figure 11). UVOT spectra show the emergence of emission around 2500–3000 Å that is later well resolved by HST/STIS as emission from Mg ii λλ2796, 2803 lines as well as Fe ii multiplets at ∼2550, 2630, 2880 Å (Figure 4). The Mg ii line profiles are similar to the H i line profiles, with a narrow component and broad, blue-shifted absorption features.

We further identify strong, narrow emission from N ii] at λλ2140, 2143 and possible emission from C iii] (λ1909) and Si iii] (λλ1892, 1896).

At shorter wavelengths, we find a mixture of high and low ionization lines (Figure 4, lower panel). We identify strong lines of C ii (λλ1334.5, 1335.7), O i (λλ1302.2–1306.0), Si ii (λλ1526.7, 1533.5). Of the higher ionization lines one notes C iv (λλ1548.2, 1550.8) and N v (λλ1238.8, 1242.8). Interestingly, N iv] λ1486.5 is either very weak or absent, which indicates a medium with density n ≳ 10⁹ cm⁻³. Fe ii is also present and Lyα emission is also very well detected.

At t ≈ t_pk + 34 days, both the optical, NIR, and UV spectra are dominated by permitted transitions. Despite the presence of high ionization lines there are no forbidden lines of, e.g., [O iii] λλ4959, 5007, N iv] λλ1486 or O iii] λλ1664, which is consistent with the picture of high density in the line forming region. The main exceptions are the [N ii] λλ2140, 2143 lines (Figure 4). The explanation could be a comparatively high critical density, i.e., ∼3 × 10⁹ cm⁻³ in combination with a high N abundance.

Interestingly, a comparison of high (C iv λλ1548.2, 1550.8 and N v λλ1238.8, 1242.8) and low (C ii λ1335) ionization emission line profiles in velocity space reveals no significant difference.

The mixture of low and high ionization lines indicates that there might be different density components or different ionization zones present in the line emitting region. The similar line profiles argue for a similar location of the ionization zones, supporting the idea of a complex emission region with different density components.

Narrow emission lines in the optical spectra of SN 2009ip require that interaction with previously ejected material (either in the form of a stable wind or from erratic mass-loss episodes) is occurring at some level. The multiple outbursts of SN 2009ip detected in the 2009, 2011 and 2012 August (from 3 yr to ∼1 month before the major 2012 explosion) are likely to have ejected conspicuous material in the immediate progenitor surroundings so that interaction of the 2012b explosion shock with this material qualifies as an efficient way to convert kinetic energy into radiation.

The radioactive decay of ⁵⁶Ni represents another obvious source of energy. We employ the nebular phase formalism developed by Valenti et al. (2008; expanding on the original work by Sutherland & Wheeler 1984 and Cappellaro et al. 1997) to constrain the amount of nickel synthesized by the 2012b explosion using late-time observations. If the observed light-curve were to be entirely powered by the energy deposition of the ⁵⁶Ni radioactive decay chain, our latest photometry would imply M_Ni  ∼  0.03 M_☉ for a standard explosion kinetic energy of E_k = 10⁵¹ erg. For a low-energy explosion with E_k = 10⁵⁰ erg, M_Ni ∼ 0.08 M_☉. Allowing for other possible sources of energy contributing to the observed luminosity (like interaction), we conclude M_Ni < 0.08 M_☉. Using this value (and the photospheric formalism by Valenti et al. 2008 based on Arnett 1982) we largely underpredict the luminosity of SN 2009ip at peak for any value of mass and kinetic energy of the ejecta: the energy release of SN 2009ip is therefore not powered by ⁵⁶Ni radioactive decay. Fraser et al. (2013) independently derived M_Ni < 0.02 M_☉, consistent with our findings. A very low ⁵⁶Ni yield (not sufficient to power the optical light-curve) was also reported by Dessart et al. (2009) and Humphreys et al. (2012) for SNe 1994W and 2011ht, respectively. Interestingly, in both cases these authors suggested a nonterminal explosion.

In the following we explore a model where the major UV-bright peak of SN 2009ip is powered by shock break-out from material ejected by the precursor bump, while continued interaction with previously ejected material is responsible for the peculiar, bumpy light-curve that follows.

The rapid rise and decay times of the major 2012b explosion (Figure 25) suggest that the shock wave is interacting with a compact shell(s) of material (see, e.g., Chevalier & Irwin 2011, their Figure 1). The relatively fast fading of CSM-like features and subsequent emergence of Type IIP features shown in Figure 21 supports a similar conclusion. We consider a model where the ejecta from the 2012b explosion initially interact with an optically thick shell of material, generating the UV-bright, major peak in the light-curve. In our model, the light-curve is powered at later times by interaction with optically thin material.

In the shock break-out scenario the escape of radiation is delayed with respect to the onset of the explosion until the radiation is able to break-out from the shell at an optical depth τ_w ≈ c/v_sh. This happens when the diffusion time t_d becomes comparable to the expansion time. Radiation is also released on the diffusion timescale, which implies that the observed bolometric light-curve rise time is t_rise ≈ t_d. Following Chevalier & Irwin (2011; see also Nakar & Sari 2010), the radiated energy at break-out E_rad, the diffusion time t_d and the radius of the contact discontinuity at t = t_d (≡ R_bo) depend on the explosion energy E, the ejecta mass M_ej, the environment density ρ_w (parameterized by the progenitor mass-loss rate) and opacity k. From our data we measure: t_rise ≈ 10 days; R_bo ≈ 5 × 10¹⁴ cm; E_rad ≈ 1.3 × 10⁴⁹ erg. We solve the system of equations for our observables in Appendix A and obtain the following estimates for the properties of the explosion and its local environment. Given the likely complexity of the SN 2009ip environment, those should be treated as order of magnitudes estimates.

The onset of the 2012b explosion is around 20 days before peak (2012 September 13). Using Equations (A2) and (A3), the progenitor mass-loss rate is $\dot{M}\approx 0.07 (v_{{\rm w}}/ 200\, {\rm km\,s^{-1}}) \,{M_{\odot }\ {\rm yr}^{-1}}$. We choose to renormalize the mass-loss rate to 200 km s⁻¹, which is the FWHM of the narrow emission component in the Hα line (e.g., Figure 16). We note that the inferred temporary mass-loss rate is appreciably below the expectations from a terminal explosion, but considerably higher than LBVs at maximum (which typically have $\dot{M}\approx 10^{-4}\hbox{--}10^{-5} \, M_{\odot }\, {\rm yr}^{-1}$; Humphreys & Davidson 1994). A similar temporary mass loss was obtained by Humphreys et al. (2012) for the Type IIn SN 2011ht in the "hot wind phase," which they interpret as a nonterminal event.

The observed bolometric luminosity goes below the level of the luminosity expected from continued interaction (Equation (A6)) around t_pk + 12 days. By this time she shock must have reached the edge of the dense wind shell: t_w ≲ 32 days. This constrains the wind shell radius to be R_w ≈ 1.2 × 10¹⁵ cm (Equation (A7)), therefore confirming the idea of a compact and dense shell of material, while the total mass in the wind shell is M_w ≈ 0.1 M_☉ (Equation (A5)).³⁹ The system of equations is degenerate for M_ej/E². Adopting our estimates of the observables above and Equation (A1) we find M_ej ≈ 50.5(E/10⁵¹ erg)² M_☉.⁴⁰ The efficiency of conversion of kinetic energy into radiation depends on the ratio of the total ejecta to wind shell mass (e.g., van Marle et al. 2010; Ginzburg & Balberg 2012; Chatzopoulos et al. 2012). This suggests M_ej ≈ M_w as order of magnitude estimate, from which E ∼ 10⁵⁰ erg.

After t_w the bolometric luminosity starts to decay faster, especially at UV wavelengths (Figure 2). Around this time the shock starts to interact with less dense, optically thin layers of material producing continued power for SN 2009ip. In this regime the observed luminosity tracks the energy deposition rate: L = 4πR²ρ_w(v_fs − v_w)³, where R is the radius of the cold dense shell that forms as a result of the loss of radiative energy from the shocked region; v_fs is the forward shock velocity; v_w and ρ_w are the velocity and density of the material encountered by the shock wave (Chevalier & Irwin 2011, and references therein). Given the complexity of the explosion environment, we adopt a simplified shock interaction model (see, e.g., Smith et al. 2010a) and parameterize the observed luminosity as: L = (η/2)wv³, where η is the efficiency of conversion of kinetic energy into radiation; $w(R)\equiv \dot{M}/ v_{{\rm w}}$ (hence ρ_w = w(R)/4πR²); while v is a measure of the expansion velocity of the shock into the environment. We use v ≈ 5000 km s⁻¹. Using the bolometric luminosity of Figure 11, we find that the total mass swept up by the shock from t_w = 32 days until the end of our monitoring (112 days since explosion) is $M_{{\rm w}}^{{\rm thin}} \approx (0.05/\eta)\ {M_{{\odot }}}$. The total mass in the environment swept up by the 2012b explosion shock is therefore $M_{{\rm tot}}=M_{{\rm w}}+M_{{\rm w}}^{{\rm thin}} \approx (0.2\hbox{--}0.3) \ {M_{{\odot }}}$ for η = 50%–30%. As a comparison, Ofek et al. (2013a) derive a mass of ∼0.1 M_☉. Our analysis points to a steep density profile with ρ_w∝R⁻⁵ for R > 1.4 × 10¹⁵ cm.

During the 2012b explosion, the shock interacts with an environment that has been previously shaped by the 2012a explosion and previous eruptions. In this section we infer the properties of the pre-2012a explosion environment, using the 2012a outburst as a probe. We look to (1) understand the origin of the compact dense shell with which the 2012b shock interacted, whether it is newly ejected material by the 2012a outburst or material originating from previous eruptions, (2) constrain the nature of the slowly moving material (v ≈ a few 100 km s⁻¹) responsible for narrow-line emission in our spectra.

We put an upper limit on the total amount of mass in the surroundings of SN 2009ip before the 2012a explosion assuming that the observed luminosity of the 2012a outburst is entirely powered by optically thin shock interaction with some previously ejected material of mass $M_{{\rm w}}^{{\rm 12a}}$. As before, L = (η/2)wv³. $M_{{\rm w}}^{{\rm 12a}}=\int w(R)dR$. The evolution of the blackbody radius of Figure 11 suggests v ≈ 2500 km s⁻¹ before 2012 August 21. We apply the same line of reasoning as above and assume that the shock continues to expand with this velocity, while the photosphere transitions to the thin regime and stalls at ≈0.4 × 10¹⁵ cm. In this picture, the 2012a shock sampled the environment on distances R < 1.2 × 10¹⁵ cm, sweeping up a total mass of $M_{{\rm w}}^{{\rm 12a}}\le (0.02/\eta)\,{M}_{\odot }$. In Section 7.2 we estimated that the total mass of the dense wind shell from which the 2012b shock breaks out is M_w ≈ 0.1 M_☉ with radius R_w ≈ 1.2 × 10¹⁵ cm. The wind shell mass is $M_{{\rm w}}= M_{{\rm w}}^{{\rm 12a}}(R<R_{{\rm w}})+M_{{\rm ej}}^{{\rm 12a}}(R<R_{{\rm w}})$, where $M_{{\rm ej}}^{12a}(R<R_{{\rm w}})$ is the portion of the ejecta mass of the 2012a explosion within R_w at t = t_w. This implies $M_{{\rm ej}}^{12a}(R<R_{{\rm w}})>0.06\,{M}_{\odot } ({>}0.09\,{M}_{\odot })$ for our fiducial efficiency η = 30% (η = 50%), comparable with the mass of the dense wind shell.

We conclude that the UV-bright 2012b explosion results from shock break-out from material ejected by the 2012a explosion, therefore establishing a direct connection between the properties of the 2012a and 2012b episodes.

The previous result also implies a solid lower limit on the total ejecta mass of the 2012a outburst: $M_{{\rm ej}}^{12a}>0.06\,{M}_{\odot } ({>}0.09\,{M}_{\odot })$ for η = 30% (η = 50%). In Section 7.2 we estimated that the total mass collected by the 2012b shock by the end of our monitoring is M_tot ≈ 0.2–0.3 M_☉, which constrains $0.06<M_{{\rm ej}}^{12a}<0.3 \,{M}_{\odot }$ for η ⩾ 30%. In the following we use $M_{{\rm ej}}^{12a}\approx 0.1\, {M}_{\odot }$ as an order of magnitude estimate for the mass ejected by the 2012a outburst.

Our spectra show evidence for narrow-line emission (Section 4) typically observed in SNe IIn (and LBVs), which is usually interpreted as signature of the ejecta interaction with material deposited by the progenitor wind before explosion. For SN 2009ip we observe during the 2012b event a velocity gradient in the narrow emission from Hα (Figure 16, panel (c)), with increasing velocity with time. This increase is consistent with being linear with time. This might suggest a Hubble-like expansion for the CSM following the simple velocity profile v∝R: as time goes by, the shock samples material at larger distances from the explosion (hence with larger velocity v). Our analysis indicates that episodes of mass ejection with approximate age 11–19 months before the 2012b explosion (roughly between 2011 February and October) might reasonably account for the observed velocity gradient. We suggest that CSM material in the surroundings of SN 2009ip moving at velocities of hundreds km s⁻¹ originates from this sudden episode(s) of mass ejection. Remarkably, SN 2009ip has been reported to be in eruptive phase between 2011 May and October (Pastorello et al. 2013), consistent with this picture.

The Hubble-like flow is not consistent with a steady wind and points instead to some mechanism leading to explosive mass ejections. Interestingly, it is during the 2011 September outburst that SN 2009ip showed evidence for material with unprecedented velocity, reaching v = 12, 500 km s⁻¹ (Pastorello et al. 2013). Since no line-driven or continuum-driven wind mechanism is known to be able to accelerate stellar surface material to these velocities (Mauerhan et al. 2013), stellar-core-related mechanisms have to be invoked. The explosive mass ejection suggested by our analysis might therefore be linked to instabilities developing deep inside the stellar core.

Davidson & Humphreys (2012) and Humphreys et al. (2012) however point out that giant eruptions driven by super-Eddington radiation might also develop a shock able to accelerate a limited amount of material to supersonic speeds. In this case, no core-related explosion would be involved. This model has been successfully applied to the Type IIn SN 2011ht (which however does not show evidence for material with v > 600 km s⁻¹; Humphreys et al. 2012) and might explain the giant eruption of η-Carinae (Davidson & Humphreys 2012). It is at the moment unclear if the same mechanism would be able to explain the very high velocities observed in the ejecta of SN 2009ip.

The analysis of Section 4.5 indicates the presence of ejecta traveling at three distinct velocities: v ∼ −12, 000 km s⁻¹, v ∼ −5500 km s⁻¹, and v ∼ −2500 km s⁻¹. These values correspond to the velocity of material seen in absorption (i.e., placed outside the photosphere). The radius of the hot blackbody R_HOT of Figure 11 tracks the position of the photosphere with time. Assuming free expansion of the ejecta and the explosion onset time (t_pk − 20 days) derived in the previous sections, we can predict at which time t_v ejecta moving at a certain velocity v will overtake the photosphere at R_HOT. Only for t ≳ t_v can the ejecta give rise to absorption features in the spectra. Spherical symmetry is an implicit assumption in the calculation of R_HOT, so that comparing the predicted t_v to the observed time of appearance of the absorption edges makes it possible to test the assumption of spherical symmetry of the explosion.

For v ∼ −2500 km s⁻¹ we find t_v = t_pk + 55 days (2012 November 27) in excellent agreement with our observations, which constrain the v ∼ −2500 km s⁻¹ absorption edge to appear between 2012 November 23 and December 5 (Figure 22). No departure from spherical symmetry needs to be invoked for slow-moving ejecta, which likely includes most of the ejecta mass.⁴¹ Spherical symmetry is instead broken by the high-velocity material traveling at v ∼ −12, 000 km s⁻¹. For the 2012b explosion we detect high-velocity material in absorption starting from ∼1 week after peak. Around peak the spectrum of SN 2009ip is optically thick and shows no evidence for material with v ∼ −12, 000 km s⁻¹ (Figures 19 and 20). However, in no way could a perfectly spherical photosphere with apparent v ≈ 4000–5000 km s⁻¹ mask the fast-moving ejecta at any time during the evolution, and in particular until the first week after peak, as we observed. This indicates a departure of the high-velocity ejecta from spherical geometry and might suggest the presence of a preferred direction in the explosion.

Asymmetry can also have a role in the spatial distribution of the interacting material, as supported by the observed co-existence of broad and (unresolved) intermediate components in the spectrum (Figure 14). In this respect, Chugai & Danziger (1994) proposed the possibility of an enhanced mass loss on the equatorial plane of SNe IIn to explain the intermediate velocity component in SN 1988Z, other explanations being a clumpy CSM or, again, an asymmetric flow. Humphreys et al. (2012), with reference to SN 2011ht, emphasized that an asymmetric outflow would strongly modify the observed spectrum. In this context it is worth noting that asymmetry is also a likely explanation for the discrepant mass-loss estimates found by Ofek et al. (2013a), as noted by the authors. A disk-like geometry for SN 2009ip was proposed by Levesque et al. (2012) based on the H Balmer decrement. Finally, the binary-star merger scenario proposed by Soker & Kashi (2013) to interpret SN 2009ip naturally leads to ejecta with a bipolar structure.

SN 2009ip is a weak X-ray and radio emitter (Figure 26). In the following two sections we connect the lack of high X-ray/radio luminosity to the shock break-out plus interaction scenario. This scenario gives rise to a two-component high-energy spectrum, with a hard (X-rays) and a soft (UV/optical, at the break-out velocity of interest here) component (e.g., Svirski et al. 2012; Chevalier & Irwin 2012). The hard component is generated by bremsstrahlung emission from hot electrons behind the shock. Theory predicts that L_{hard, bo} ∼ 10⁻⁴ L_bo (where L_bo is the break-out luminosity, resulting from the soft and hard component) as long as: (1) inverse Compton (IC) cooling dominates over bremsstrahlung; (2) high-energy photons undergo Compton degradation in the unshocked wind during their diffusion to the observer. We show in the following that both processes are relevant for SN 2009ip and provide a natural explanation for its very low X-ray to optical luminosity ratio (L_x/L_optical ≲ 10⁻⁴).

Zoom In Zoom Out Reset image size

From our modeling of Section 7.2, we inferred a density parameter D_* ≈ 0.4 (see Appendix A for a definition of D_*). With a shock velocity v ∼ 4500 km s⁻¹ (or larger), the density measurement above implies that around break-out time the main source of cooling of the hot electrons is IC (see Chevalier & Irwin 2012). For our parameters, the bremsstrahlung (ff) to IC emissivity ratio at break-out is (Svirski et al. 2012, their Equation (17)) ^ff/^IC ≈ 0.01 (v/10⁹ cm s⁻¹)⁻² or ^ff/^IC ≈ 0.05. IC is the dominant cooling source, suppressing the emission of hard photons in SN 2009ip. The calculations by Ofek et al. (2013a) instead assume negligible IC cooling.

Comptonization of the hard photons as they propagate through the unshocked wind region to the observer furthermore leads to a suppression of high-energy radiation. This process can effectively suppress photons with ≈keV energy if τ_es ≳ 15–20, the photon energy being limited by $\epsilon _{{\rm max}}=511/\tau _{{\rm es}}^2$ keV. Our modeling of Section 7.2 implies τ_es ≈ 15 for SN 2009ip around break-out. This demonstrates that both the domination of IC over bremsstrahlung (1) and Compton losses (2) are relevant to explain the weak X-ray emission in SN 2009ip. Identifying L_x with L_hard and L_bol with L_bo, the shock break-out scenario therefore naturally accounts for the observed L_x/L_optical ≲ 10⁻⁴ ratio even in the absence of photo-absorption.

Chevalier & Irwin (2012) calculate that full ionization (which gives minimal photo-absorption) is achieved only for high-velocity shocks with v ≳ 10⁴ km s⁻¹. Slower shocks would lead to incomplete ionization (i.e., potentially important photo-absorption), which will further reduce the escaping X-ray flux. Using the explosion observables and Equation (A8) we constrain the total neutral column density between the shock break-out radius and the observer to be N ≈ 2 × 10²⁵ cm⁻², which gives a bound–free optical depth τ_bf ≈ 2 × 10³ at 1 keV (Equation (A9)). Since τ_bf∝R⁻¹ and R∝t, soft (E ≈ 2 keV) X-ray emission would not be expected from SN 2009ip until R ≈ 2 × 10¹⁶ cm, which happens for Δt > 220 days since the explosion, or >20 break-out timescales t_d (where t_d ≈ t_rise in our scenario).⁴² This calculation assumes an extended and spherically symmetric wind profile. The Chandra detection of SN 2009ip at much earlier epochs (Δt ≈ 4 t_d) indicates that at least one of these assumptions is invalid, therefore pointing to a truncated and/or highly asymmetric wind profile. This is consistent with the picture of a dense but compact wind shell of radius R_w ≈ 1.2 × 10¹⁵ cm followed by a steep density gradient ρ_w∝R^−5.3 we developed in Section 7.2. Asymmetry also plays a role, as independently suggested by observations in the optical (Section 7.4).

Finally, Katz et al. (2011) predict that hard-X-ray emission with typical energy of 60 keV is also produced by the collisionless shock. The details of the spectrum are however unclear. Bound–free absorption is less important at these energies giving the chance to detect hard X-rays at earlier times. Our Swift-BAT campaign in the 15–150 keV range revealed a tentative detection. With these observations we put a solid upper limit on the hard X-ray- to- optical luminosity around maximum light, which is L_{X, hard}/L_bol < 5 × 10⁻³ at 5σ. The broadband SED around maximum light is shown in Figure 27.

Zoom In Zoom Out Reset image size

The shock-CSM interaction is a well-known source of radio emission (e.g., Chevalier 1982, 1984). The high density of the wind shell derived from our modeling of Section 7.2 implies a free–free optical depth at radius R, τ_ff ≈ 3.8 × 10⁵⁴(ν/GHz)^−2.1(R/cm)⁻³ (Equation (A10)). With τ_ff ≈ 10⁵–10⁸ at R ≈ R_bo, no detectable radio emission is expected around break-out, consistent with our lack of radio detection around these times. If the dense wind profile extends out to large distances, the calculation above shows that no radio emission is expected until very late times, when the shock reaches R ≈ 7 × 10¹⁶ cm. Our radio detection at much earlier epochs (Δt ≈ 60–80 days since explosion) demonstrates instead that the dense wind shell is not extended but truncated and adds further, independent evidence for a complex medium where inhomogeneity, asymmetry and/or low wind filling factor might also be relevant. This is consistent with the idea we suggested in Section 7.3 that the dense shell was the outcome of a short eruption, since eruptions are more likely to eject shells as opposed to a steady mass loss.

The collision of the ejecta with massive, dense shells is expected to accelerate cosmic rays (CRs) and generate GeV gamma-rays (Murase et al. 2011; Katz et al. 2011) with fluence that depends both on the explosion and on the environment parameters. The GeV emission is expected to be almost simultaneous with the UV–optical–NIR emission.

The proximity of SN 2009ip (≈24 Mpc) justifies the first search for GeV emission from shock break-out. Fermi-LAT observations covering the period t_pk − 30 days until t_pk + 28 days yielded no detection (Section 2.7). Following Murase et al. (2011), we predict the GeV emission from SN 2009ip (2012b outburst) using the explosion and environment parameters inferred in Section 7.2 (shell density ρ_w, corresponding to n ≈ 4 × 10¹⁰ cm⁻³ at the break-out radius, or mass M_w) together with the observables of the system (t_rise, R_bo). We take into account the attenuation due to γγ → e⁺e⁻ pair production, assuming a blackbody spectrum with T = 20, 000 K. The attenuation by extragalactic background light is also included. Since n is not too large, the Bethe–Heitler pair production will be irrelevant in our case, as it happens for a clumpy or asymmetric CSM (Section 7.4). The injected CR energy E_CR is assumed to be equal to that of the radiation energy at t_rise. We compare the expected GeV fluence with observations in Figure 27: the Fermi-LAT nondetection is consistent with our picture of ejecta crashing into a compact and dense but low-mass (≲ 0.1–0.2 M_☉) shell of material at small radius that we developed in the previous sections. For the detection of γ-rays, brighter SNe (closer SNe or SNe accompanying larger dissipation) are needed.

Using the set of parameters above we predict the muon and antimuon neutrino fluence from SN 2009ip in Figure 28, using the model by Murase et al. (2011). CRs produce mesons through inelastic proton–proton scattering, leading to neutrinos as well as γ-rays. As for the γ-rays, given that the explosion/ environment parameters are constrained by observations at UV/optical/NIR wavelengths, the neutrino fluence directly depends on the total energy in CRs, E_CR (or, equivalently on the CR acceleration efficiency _CR, E_CR ≡ _CRE, being E the explosion energy). Note that the injected CR energy E_CR is assumed to be the radiation energy at t_rise, but larger values of E_CR are possible. The maximum proton energy is also set to 1.5 PeV, corresponding to ε_B = 0.01. Our calculations show that SN 2009ip was a bright source of neutrinos if E_CR is comparable to the radiated energy. We note however that its location in the southern sky was not optimal for searches for neutrino emission by IceCube because of the severe atmospheric muon background.

Zoom In Zoom Out Reset image size

With this study we demonstrate that it is possible to constrain the CR acceleration efficiency if: (1) the explosion/ environment parameters are constrained by observations at UV/optical/NIR wavelengths; (2) deep limits on muon neutrinos are available, as will be the case also for southern sky sources once KM3Net will be online in the near future.

The time-resolved broadband SED analysis of Section 3 identifies the presence of a NIR excess of emission (Figure 13) with respect to a blackbody spectrum. Contemporaneous NIR spectroscopy (Section 2.3) shows that the NIR excess cannot be ascribed to line emission (Figure 7), therefore pointing to a physical process producing NIR continuum emission with luminosity L ∼ 4 × 10⁴¹ erg s⁻¹. Here we discuss four different physical scenarios: (1) dust formation, (2) light echo from dust, (3) dust vaporization, and (4) radiation from gas in an extended outflow. The last of these seems most likely.

We first adopt a blackbody spectral model to obtain rough estimates of the temperature and location of the dust and constrain the dust-related scenarios. In a more realistic dust emission model, we would need to consider the dust grain properties and composition, leading to a spectrum that is not a true blackbody (e.g., Fox et al. 2011, 2013). This is material for a dedicated article (R. Margutti et al., in preparation). Here we use a simplified approach. From the SED analysis of Section 3 we find that the equivalent NIR blackbody radius is R_COLD ∼ 4 × 10¹⁵ cm with very limited evolution with time (Figure 11). The cold blackbody temperature is also stable, with T_COLD ∼ 3000 K. Considering that our simplified approach overestimates the true dust temperature of hundreds of degrees (up to 20% according to Nozawa et al. 2008) the real dust temperature should be closer to ∼2500 K.

A clear signature of dust formation is the development of highly asymmetric and blue-shifted line profiles (see, e.g., Smith et al. 2008, their Figure 4), which is not observed. The temperature of ∼2500–3000 K is also prohibitively high for dust to form. At this very early epochs the shock radius is also <R_COLD, implying that the NIR emission cannot originate from dust created behind the reverse shock. The dust creation scenario is therefore highly unlikely, leading to the conclusion that the NIR excess originates from preexisting material ejected by the progenitor before the 2012b explosion. Since the geometry of the NIR emitting material can be non spherical (Section 7.4) and/or have low filling factor (i.e., the material can be clumpy), we do not expect this material to necessarily produce absorption along our line of sight.

Our SED fitting indicates the presence of a NIR excess with similar radius during the 2012a eruption as well (Figure 11), suggesting that the origin of the preexisting material is rather linked to the eruption episodes of the progenitor of SN 2009ip in the years before. The same conclusion was independently reached by Smith et al. (2013). We note that the size of the cool emitting region of SN 2009ip is remarkably similar to the preoutburst dust radii of other optical transients linked to eruption episodes of their progenitor stars like NGC 300 OT2008-1 (R ∼ 5 × 10¹⁵ cm) and SN 2008S (R ∼ 2 × 10¹⁵ cm; seem e.g., Berger et al. 2009b, their Figure 28).

A light echo from dust (i.e., preexisting dust heated up by the UV and optical radiation from the explosion) would require the dust grains to survive the harsh environment (see however Smith et al. 2013). At the high temperature of T ∼ 2500–3000 K this is however unlikely, while dust vaporization is more likely to happen (see, e.g., Draine & Salpeter 1979). We speculate on the dust vaporization scenario below.

In this picture a cavity is excavated by the explosion radiation out to a radius R_c: this radius identifies the position of the vaporized dust, while it does not track the outer dust shell radius (which is instead likely to expand with time; see, e.g., Pearce et al. 1992). Being the dust shell created in the years before the 2012 explosion, we expect the smaller grains to be located at the outer edge of the dust shell as a result of forces acting on them (e.g., radiative pressure). At smaller distances we are likely dominated by the larger dust grains. Following Dwek et al. (1983), their Equation (8) (see also Draine & Salpeter 1979) the radius of the dust-free cavity for a UV–optical source with luminosity L is: R_c = 23(〈Q〉(L/L_☉)/(λ₀/μm)T⁵)^0.5, where R_c is in units of pc; T is the temperature at the inner boundary of the dust shell that we identify with the evaporation temperature T_ev; 〈Q〉 is the grain emissivity: Q = (λ₀/λ)ⁿ for λ ⩾ λ₀ while Q = 1 for λ < λ₀. Here we adopt n = 1.⁴³ The dust grain radius is a, with λ₀ ∼ 2a, which implies R_c∝a^−0.5. For R_c = R_COLD ∼ 4 × 10¹⁵ cm, T_ev ∼ 3000 K and L = L_pk ∼ 1.7 × 10⁴³ erg s⁻¹, we constrain the radius of the vaporizing dust grains to be a ≲ (2–5) μm.

Grain radii of 0.2–2 μm are typically found in dust shells, suggesting that we are possibly witnessing the vaporization of the larger grains at R ∼ R_c. The very high vaporization temperature is only potentially compatible with materials like graphite, silicates, corundum and carbide, which have binding energy U₀ ≳ 6 eV. Even in these cases T_ev ∼ 2500–3000 K requires extremely short vaporization timescales of the order of ∼1 day or less and large grains of the order of 1 μm (Draine & Salpeter 1979, their Equation (24)).

A more natural possibility is that of free–free emission from a power-law distribution of material at large radii (i.e., an extended outflow), ionized by SN 2009ip (Gall et al. 2012). Free–free emission is well known to produce an excess of IR emission in hot stars surrounded by dense winds (see, e.g., Wright & Barlow 1975). Free–free emission has been ruled out by Smith et al. (2013) as potential explanation of the NIR excess of SN 2009ip based on a mismatch between the predicted and observed slope of the continuum. However, the free–free spectral index depends on the density distribution ρ(R) of the emitting material: for ρ(R)∝R^−β, F_λ∝λ^{−(8β − 8.2)/(2β − 1)} (Wright & Barlow 1975). The mild spectral index of the NIR excess (see Figure 13) implies a mild density gradient at large radii, with β < 2 (i.e., significantly flatter than the standard wind profile expected in the case of constant mass-loss rate). This finding suggests a significant increase in the mass-loss rate of the progenitor star during the years preceding the 2012 double outburst.

SN 2010mc (Ofek et al. 2013b) is the only other hydrogen-rich explosion with clear signs of interaction and a detected precursor. The similarity with SN 2009ip extends to the energetics, timescales (Figure 29), and spectral properties both during the precursor and the major rebrightening, as noted by Smith et al. (2013).

Zoom In Zoom Out Reset image size

The first conclusion is that (1) the precursor and the major rebrightening are causally connected events, being otherwise difficult to explain the strict similarity between two distinct and unrelated explosions. The same conclusion is independently supported by the very short time interval between the precursor and main outburst when compared to the progenitor star lifetime, as pointed out by Ofek et al. (2013b) for SN 2010mc. A second conclusion is that (2) whatever causes the precursor plus major outburst phenomenology, this is not unique to SN 2009ip and might represent an important evolutionary channel for massive stars.

Furthermore, (3) SNe 2009ip and 2010mc must share some fundamental properties. Their evolution through the explosive phase must be driven by few physical parameters. A more complicated scenario would require unrealistic fine tuning to reproduce their close similarity. This also suggests we are sampling some fundamental step in the stellar evolution of the progenitor system. Finally, (4) whatever the physical mechanism behind, the time interval of ∼40 days (Figure 29) between the precursor explosion and the main event must be connected to some physically important timescale for the system.

We employ the fast χ² technique for irregularly sampled data by Palmer (2009) to search for periodicity and/or dominant timescales in the outburst history of SN 2009ip before⁴⁴ the major 2012 explosion. Details can be found in Appendix B. Applying the method above to the R-band data we find evidence for a dominant timescale of ∼38 days, (with significant power distributed on timescales between 30 and 50 days), intriguingly similar to the Δt ∼ 40 days between the precursor and major explosion in 2012. We emphasize that this is not a claim for periodicity, but the identification of a dominant variability timescale of the signal.

Our analysis identifies the presence of a fundamental timescale that regulates both the progenitor outburst history and the major explosion, and that is also shared by completely independent events like SN 2010mc. This timescale corresponds to a tiny fraction of a massive star lifetime: ∼10⁻⁸ for τ ∼ 4–6 Myr, as appropriate for a 45–85 M_☉ star. We speculate on the nature of the underlying physical process in the next two sections.

Asymmetry likely plays a role in the 2012 explosion (Section 7.4), which might point to the presence of a preferred direction in the progenitor system of SN 2009ip. This suggests either a (rapidly?) rotating single star or an interacting binary as progenitor for SN 2009ip. We first update previous estimates of the progenitor mass of SN 2009ip using the latest stellar evolutionary tracks and then discuss the effects of stellar rotation and binarity.

From HST preexplosion images, employing the latest Geneva stellar evolutionary tracks (Ekström et al. 2012) that include important updates on the initial abundances, reaction rates and mass-loss prescriptions with respect to Schaller et al. (1992), we determine for M_V = −10.3 ± 0.3 (Foley et al. 2011) a zero-age main-sequence (ZAMS) mass M ≳ 60 M_☉ assuming a nonrotating progenitor at solar composition, consistent with previous estimates. M_V = −10.3 corresponds to L_V ∼ 10^6.1 L_☉. This implies L_bol > 2 × 10⁶ L_☉, thus rivaling the most luminous stars ever discovered (e.g., Crowther et al. 2010). Luminosities of a few 10⁶ L_☉ have been indeed associated to the group of LBV stars with typical temperature of ∼15, 000–25, 000 K. Adopting this range of temperature results in L_bol ∼ 5 × 10⁶ L_☉, suggesting that any progenitor with M < 160 M_☉ was super Eddington at the time of the HST observations. For (2 < L_bol < 5) × 10⁶ L_☉ the allowed mass range is 60 M_☉ < M_ZAMS < 300 M_☉ (e.g., Crowther et al. 2010). Including the effects of axial rotation results in a more constrained range of allowed progenitor mass: 45 M_☉ < M_ZAMS < 85 M_☉ (for Ω/Ω_crit = 0.4).

Rapid rotation strongly affects the evolution of massive stars, in particular by increasing the global mass-loss rate (e.g., Maeder & Meynet 2000). More importantly, Maeder (2002) showed that the mass loss does not remain isotropic, but it is instead enhanced at the polar regions, thus favoring bipolar stellar winds (e.g., Georgy et al. 2011, their Figure 2). As a result, the formation of an asymmetric (peanut shaped) nebula around rapidly rotating stars is very likely. Additionally, rapid rotation can induce mechanical mass loss, resulting in some matter to be launched into an equatorial Keplerian disk. It is clear that any explosion/eruption of the central star will thus naturally occur in a nonisotropic medium, as we find for SN 2009ip. Rotation further leads to enhanced chemical mixing (e.g., Chatzopoulos et al. 2012; Yoon et al. 2012).

HST preexplosion images cannot, however, exclude the presence of a compact companion.⁴⁵ The massive progenitor of SN 2009ip might be part of a binary system and the repetitive episodes of eruption might be linked to the presence of an interacting companion. The close periastron passages of a companion star in an eccentric binary system has been invoked by Smith & Frew (2011) to explain the brightening episodes of ηCar in 1838 and 1843. In the context of SN 2009ip, Levesque et al. (2012) discussed the presence of a binary companion, while Soker & Kashi (2013) suggested that the 2012b explosion was the result of a merger of two stars, an extreme case of binary interaction. A binary scenario was also proposed by Soker (2013) for SN 2010mc.

A binary system would have a natural asymmetry (i.e., the preferred direction defined by the orbital plane) and a natural timescale (i.e., the orbital period) as indicated by the observations. A possible configuration suggested to lead to substantial mass loss is that of a binary system made of a compact object (neutron star) closely orbiting around a massive, extended star (Chevalier 2012 and references therein). In this picture the mass loss is driven by the inspiral of the compact object in the common envelope (CE) evolution and the expansion velocity of the material is expected to be comparable to the escape velocity for the massive star. For⁴⁶ v ∼ 1000 km s⁻¹ and M ∼ 45–85 M_☉ the radius of the extended star is R_* < (1–2) × 10¹² cm (where the inequality accounts for the fact that we used the ZAMS mass). In this scenario, the mass loss is concentrated on the orbital plane of the binary (Ricker & Taam 2012), offering a natural explanation for the observed asymmetry. However, it remains unexplained how this mechanism would be able to launch material with v ∼ 10⁴ km s⁻¹ as observed during the 2012a episode. A potential solution might be the presence of a high-velocity accretion disk wind.

While the presence of a dominant timescale common to the eruptive and explosive phases makes the binary progenitor explanation particularly appealing, the CE model in its present formulation seems to have difficulties in explaining the observed phenomenology. Alternative scenarios are explored in Section 8.3.

The outer envelope of an evolved massive star contains ∼95% of the stellar radius but only a tiny fraction of the stellar mass. The physical mechanism leading to the 2012a outburst must have been fairly deep seated to unbind more than ∼0.1 M_☉. It is furthermore required to generate a large amount of energy (10⁴⁸ erg plus the energy to lift the mass out of the deep gravitational potential well) and to explain the presence of a dominant timescale of ∼40 days, which is also shared with the observed eruption history in the years before the explosion. The mass-loss rate of at least a few ∼0.1 M_☉ yr⁻¹ can only be sustained for an extremely small fraction of the life of a star and in principle only for ∼ years, before inducing important adjustments to the stellar structure.

We explore here different physical mechanisms that have been proposed to lead to substantial mass loss in the late stages of evolution of massive single stars and discuss their relevance for SN 2009ip. They can basically be grouped into four categories: (1) pulsational pair instability (PPI), (2) super-Eddington fusion luminosity, (3) Eddington-limit instabilities, and (4) shock heating of the stellar envelope.

PPI (Barkat et al. 1967) applies to massive stars with helium core of ∼40–60 M_☉. PPI leads to partial unbinding of the star, with a sequence of eruptions accompanied by the ejection of shells of material of the stellar envelope. This mechanism has been considered as a plausible explanation for SN 2009ip by Pastorello et al. (2013), Mauerhan et al. (2013) and Fraser et al. (2013) but has been rejected by Ofek et al. (2013b) and Ofek et al. (2013a) for SNe 2010mc and 2009ip (the leading argument being that PPI would lead to the ejection of much larger amounts of mass than the ∼10⁻¹–10⁻² M_☉ they infer). According to Woosley et al. (2007), a nonrotating star with zero metallicity (Z = 0) and ZAMS mass 95 and 130 M_☉ meets the criteria for PPI. However, with updated prescriptions for the mass-loss rate, and assumption of solar metallicity (which more closely represents our conditions of 0.4 Z_☉ < Z < 0.9 Z_☉, Section 5) Ekström et al. (2012) predict that even stars with M = 120 M_☉ will end the C-burning phase with mass ∼31 M_☉, below the threshold of ∼40 M_☉ to trigger PPI. While our limits on the progenitor mass of SN 2009ip in Section 8.2 do not formally rule out the PPI scenario in the nonrotating case, they definitely allow for progenitors starting with much lower mass than required for the PPI to develop (M ≳ 120 M_☉).

Rapidly rotating progenitors (Ω/Ω_crit > 0.5) enter the PPI regime starting with substantially lower ZAMS mass: ∼50 M_☉ for Z = 0 (Chatzopoulos & Wheeler 2012; Yoon et al. 2012). On the other hand, for Z = Z_☉ even very massive rotating 120 M_☉ stars develop a core with M ∼ 19 M_☉, insufficient to trigger PPI (Ekström et al. 2012). In the previous sections we constrained the progenitor of SN 2009ip with ZAMS mass 45 M_☉ < M < 85 M_☉ and metallicity 0.4 Z_☉ < Z < 0.9 Z_☉. Following the prescriptions from Chatzopoulos & Wheeler (2012), adopting the most favorable conditions (i.e., M = 85 M_☉ and Z = 0.4 Z_☉) and starting with a very rapidly rotating star with Ω/Ω_crit = 0.9, we find the final oxygen core mass to have M ∼ 35 M_☉, formally below the threshold of 40 M_☉ for PPI. This indicates some difficulties for rotating progenitors to reach the PPI threshold. What remains difficult to interpret both for rotating and nonrotating progenitors is the presence of a dominant timescale of ∼40 days: depending on the pulse properties and the physical conditions of the surviving star, the interval between pulses can be anywhere between days to decades (Woosley et al. 2007).

Alternatively, convective motions stimulated by the super-Eddington fusion luminosity in the core of a massive star can excite internal gravity waves (g-mode nonradial oscillations; Quataert & Shiode 2012; Shiode & Quataert 2013). An important fraction of the energy in gravity waves can be converted into sound waves, with the potential to unbind several M_☉. For a 40 M_☉ star, Quataert & Shiode (2012) estimate that 10⁴⁷–10⁴⁸ erg will be deposited into the stellar envelope. If this mechanism is responsible for the ejection of material by the 2012a explosion, the velocity v ∼ 1000 km s⁻¹ of the unbound material inferred from our observations implies an ejecta mass ≲ 0.1 M_☉, consistent with our estimate of the mass of the compact shell encountered by the 2012b explosion in Section 7.2. However, this mechanism is likely to lead to a steady mass loss (E. Quataert 2013, private communication) as opposed to shell ejection and does not offer a natural explanation for the 40-day timescale involved in the process. We therefore consider the super-Eddington luminosity mechanism unlikely to apply to SN 2009ip and SN 2010mc (see, however, Ofek et al. 2013b).

With R-band observed magnitudes between ∼18 mag and ∼21 mag (Pastorello et al. 2013), SN 2009ip oscillates between ∼Eddington and super-Eddington luminosity in the years preceding the double 2012 explosion. Depending on the effective temperature of the emission, R ∼ 21 mag corresponds to the Eddington luminosity of a star with mass 40–80 M_☉. R ∼ 18 would correspond to an Eddington limit for a star of M = 600–1200 M_☉. A number of instabilities have been shown to develop in stars approaching and/or exceeding the Eddington limit, allowing the super-Eddington luminosity to persist (e.g., Owocki & Shaviv 2012), potentially powering LBV-like eruptions. The eruption timescale, repetition rate and ejected mass are however loosely predicted, so that it is unclear if any of these mechanisms would apply to the eruption history of SN 2009ip and, even more importantly, how these are connected with the 2012 double explosion (which clearly differs from super-Eddington powered winds). The inferred mass-loss rate of $\dot{M}\sim 0.07 \,{M_{\odot }\ {\rm yr}^{-1}}$ is appreciably higher than the typical mass loss of $\dot{M}\sim 10^{-4}\hbox{--}10^{-5} \,{M_{\odot }\ {\rm yr}^{-1}}$ associated with classical LBVs (Humphreys & Davidson 1994). This finding suggests that the true evolutionary state of the progenitor of SN 2009ip might be different from a classical LBV. With luminosity between ∼10⁶ and ∼10⁷ L_☉, the progenitor of SN 2009ip falls into the region where radiation pressure starts to have a major role in supporting the star against gravity (e.g., Owocki & Shaviv 2012, their Figure 12.1): we speculate that the dominance of radiation pressure over gas pressure in the envelope might have an important role in determining the repeated outbursts of SN 2009ip.

The outer layers of massive stars close to the Eddington limit and/or critical rotation are only loosely bound and might be easily ejected if enough energy is suddenly deposited. A potential source of energy deposition has been identified in thermonuclear flashes associated with shell burning (Dessart et al. 2010). Differently from the PPI, this scenario does not require the progenitor to be extremely massive. Alternatively, nonradial gravity mode oscillations above the core or near to the H-burning shell could provide fresh fuel, triggering a burst of energy (and subsequent shell ejection) in massive stars like ηCar (Guzik 2005). Both scenarios have the advantage to be basically driven by two parameters: the deposited energy and the envelope binding energy, naturally satisfying the "simplicity" criterion of Section 8.1.

Finally we mention that direct numerical simulations of precollapse hydrodynamics by Meakin (2006), Meakin & Arnett (2007) and Arnett & Meakin (2011a, 2011b) found eruptive instabilities due to the interaction of oxygen and silicon burning shells. The instabilities are related to turbulent convection (and being inherently nonlinear, are invisible to conventional linear stability analysis). These simulations specifically predicted mass ejection prior to core collapse, suggesting that precollapse evolution is far more eventful than previously thought. However, as for all the other mechanisms analyzed in this section, it is at the moment unclear how to explain the 40 day timescale.

Our modeling shows that the 2012b explosion is not powered by Ni radioactive decay (Section 7.1) and demonstrates that a shock breaking out from material previously ejected by the 2012a outburst can reasonably account for the observed properties of the 2012b explosion, constraining the mass of the ejecta to M_ej = 50.5(E/10⁵¹ erg)² M_☉ (Section 7.2). This strongly suggests an explosion energy below 10⁵¹ erg (likely around 10⁵⁰ erg), and brings to question the fate of SN 2009ip: a terminal SN explosion (Mauerhan et al. 2013) or a SN "impostor" (Pastorello et al. 2013; Fraser et al. 2013)?

A key prediction of the SN scenario is the presence of chemical elements produced by the SN nucleosynthesis (like oxygen) in late-time spectra, which is not expected in the case of a nonterminal explosion. Our latest spectrum acquired on 2013 April 11 (t_pk + 190 days) still shows no evidence for SN-synthesized material (Figure 30). This nondetection is consistent with (but surely not a proof of) the nonterminal explosion scenario. This spectrum, dominated by H emission lines and where the high-velocity absorption features have completely disappeared, shares some similarities with preexplosion spectra of SN 2009ip. A notable difference, however, is the presence of (intermediate width) He i and [Ca ii]. The present data set therefore does not yet offer compelling evidence for a standard (i.e., resulting from the collapse of a degenerate core) SN-explosion scenario.

Zoom In Zoom Out Reset image size

Interaction seems to be still dominating the emission at this time. Our late-time observations indicate a flattening in the light-curve (our latest V-band photometry acquired at t_pk + 190 days implies a decay of ΔV = +0.86 ± 0.22 over 98.5 days, or 0.0087 ± 0.0022 mag day⁻¹), which might be caused by continued interaction of the shock with the NIR emitting material.

We propose that SN 2009ip was the consequence of the explosive ejection of the envelope of a massive star, which later interacted with material ejected during previous eruption episodes. While E < 10⁵¹ erg is insufficient to fully unbind a massive star, its outer envelope is only loosely bound, and can be easily ejected by a lower-energy explosion. The origin of the energy deposition is not constrained as unclear is its potential relation with the possible binary nature of the system. The final fate of SN 2009ip depends on the properties of the remaining "core," and in particular, on its mass and rotation rate: if the star managed to explode its entire H envelope still retaining a super-critical core mass, it might reexplode in the future as a genuine H-poor (Ib/c) SN explosion; if instead the star partially retained its envelope, it will possibly give rise to other SN-impostor displays, on timescales that are difficult to predict. The core might also have directly collapsed to a black hole: in this case this would mark the "end." Only close inspection of the explosion site after SN 2009ip has appreciably faded will reveal if the star survived the ejection of its outer layers. Given the impressive similarity with SN 2010mc, our interpretation extends to this event as well.

Swift-UVOT photometry in the six filters has been analyzed following the prescriptions by Brown et al. (2009). We used different apertures during the fading of the 2012a outburst to maximize the signal-to-noise ratio and limit the contamination by a nearby star. For the 2012a event we used a 3'' aperture for the b and v filters, 4'' for the u filter, and 5'' for the UV filters. We correct for point spread function (PSF) losses following standard prescriptions. At peak SN 2009ip reaches u ∼ 12.5 mag potentially at risk for major coincidence losses. For this reason we requested a smaller readout region around maximum light. Our final photometry is reported in Table 4 and shown in Figure 2. The photometry is based on the UVOT photometric system (Poole et al. 2008) and the revised zero points of Breeveld et al. (2011).

Starting on 2012 September 27 (t_pk − 6 days), a series of spectra were taken with the Swift-UVOT UV grism, with a cadence of one to two days, until 2012 October 28 (t_pk + 25 days), with a final long observation on 2012 November 2 (t_pk + 30 days). The spectra were extracted from the image using the UVOTPY package. The wavelength anchor and flux calibration were the recent updates valid for locations other than the default position in the center (P. Kuin et al., in preparation; details can be found online⁴⁸). The spectra were extracted for a slit with the default 2.5σ aperture and a 1σ aperture, applying an aperture correction when needed. The wavelength accuracy is 20 Å (1σ), the flux accuracy (systematic) is within 10%, while the resolution is R ∼ 75–110 depending on the wavelength range. The error in the flux was computed from the Poisson noise in the data, as well as from the consistency of between the spectra extracted from the images on one day. The sequence of Swift-UVOT spectra is shown in Figure 3. The observing log can be found in Table 1.

Observations with the Space Telescope Imaging Spectrograph (STIS) were taken on 2012 October 29 (t_pk + 26 days) using aperture 52x0.2E1 with gratings G230LB, G430L, and G750L with exposures times of 1200 s, 400 s, and 100 s, respectively. The STIS two-dimensional images were cleaned of cosmic rays and dead pixel signatures before extraction. The extracted spectra were then matched in flux to the STSDAS/STIS pipeline one-dimensional data product. The spectrum is shown in Figure 4. Further HST-COS data were acquired on 2012 November 6 (t_pk + 34 days, Figure 4, lower panel). Observations with the Cosmic Origin Spectrograph (COS) were acquired using MIRROR A + bright object aperture for 250 s. The COS data were then reprocessed with the COS calibration pipeline, CALCOS v2.13.6, and combined with the custom IDL coaddition procedure described by Danforth et al. (2010) and Shull et al. (2010). The total exposure time for the far-UV observation was 3100 s and 3700 s for near-UV.

Observations in the v, b, and u filters were obtained with Swift-UVOT and reduced as explained above. In Figure 2 we apply a dynamical color term correction to the UVOT v, b, and u filters to plot the equivalent Johnson magnitudes following the prescriptions by Poole et al. (2008). This is a minor correction to the measured magnitudes, and it is not responsible for the observed light-curve bumps. R- and I-band photometry was obtained with the UIS Barber Observatory 20 inch telescope (Pleasant Plains, IL), operated by Josch Hambasch at the Remote Observatory Atacama Desert, a Celestron C9.25 operated by TG Tan (Perth, Australia), and a C11 Schmidt-Cassegrain telescope operated by Ivan Curtis (Adelaide, Australia). Exposure times ranged from 120 s to 600 s. Images were reduced following standard procedure. The photometry was corrected to the photometric system of Pastorello et al. (2013) using the corrections of dR = +0.046 and dI = +0.023 (A. Pastorello 2012, private communication). R- and I-band photometry is reported in Table 5. A single, late-time (t_pk + 190 days) V-band observation was obtained with the Inamori-Magellan Areal Camera and Spectrograph (IMACS; Dressler et al. 2006) mounted on the Magellan/Baade 6.5 m telescope on 2013 April 11.40 (exposure time of 90 s). Data were reduced using standard tasks in IRAF⁴⁹ and calibrated to a standard star field at similar airmass.

SN 2009ip was observed with the MagE (Magellan Echellette) Spectrograph mounted on the 6.5 m Magellan/Clay telescope at Las Campanas Observatory. Data reduction was performed using a combination of Jack Baldwin's mtools package and IRAF echelle tasks, as described in Massey et al. (2012). Optical spectra were obtained at the F. L. Whipple Observatory 1.5 m Tillinghast telescope on several epochs using the FAST spectrograph (Fabricant et al. 1998). Data were reduced using a combination of standard IRAF and custom IDL procedures (Matheson et al. 2005). Low- and medium-resolution spectroscopy was obtained with the Robert Stobie Spectrograph mounted on the Southern African Large Telescope (SALT/RSS) at the South African Astronomical Observatory in Sutherland, South Africa. Additional spectroscopy was acquired with the Goodman High Throughput Spectrograph (GHTS) on the SOAR telescope. We also used IMACS mounted on the Magellan telescope and the Low Dispersion Survey Spectrograph 3 (LDSS3) on the Clay telescope (Magellan II). The Multiple Mirror Telescope (MMT) equipped with the "Blue Channel" spectrograph (Schmidt et al. 1989) was used to monitor the spectral evolution of SN 2009ip over several epochs. Further optical spectroscopy was obtained with the R-C CCD Spectrograph (RCSpec) mounted on the Mayall 4.0 m telescope, a Kitt Peak National Observatory (KPNO) facility. Spectra were extracted and calibrated following standard procedures using IRAF routines. We used the X-shooter echelle spectrograph (D'Odorico et al. 2006) mounted at the Cassegrain focus of the Kueyen unit of the Very Large Telescope (VLT) at the European Southern Observatory (ESO) on Cerro Paranal (Chile) to obtain broadband, high-resolution spectroscopy of SN 2009ip on 2012 September 30 (t_pk − 3 days) and October 31 (t_pk + 28 days). The spectra were simultaneously observed in three different arms, covering the wavelength range 3000–25000 Å. We used the X-shooter pipeline (Modigliani et al. 2010) and custom IDL programs to reduce and flux calibrate the data. The spectra were also slit-loss corrected and corrected for heliocentric velocities. The spectra were not (carefully) telluric corrected. The sequence of optical spectra is shown in Figure 6. The observing log can be found in Table 2.

UKIRT observations in the ZYJHK filters started on 2012 September 23 (t_pk − 10 days) and continued on a nearly daily basis until 2012 December 31 (t_pk + 89 days), when SN 2009ip settled behind the Sun. The flux of the object was measured with AUTO-MAG of SExtractor (Bertin & Arnouts 1996), where the photometric calibration was done using Two Micron All Sky Survey (2MASS) stars within a radius of 8 arcmin from SN 2009ip. The 2MASS magnitudes of the stars were converted to the UKIRT system following Hodgkin et al. (2009). Additional NIR photometry was obtained with PAIRITEL, the f/13.5 1.3 m Peters Automated Infrared Imaging TELescope at the Fred Lawrence Whipple Observatory (FLWO) on Mount Hopkins, Arizona (Bloom et al. 2006). PAIRITEL data were processed following Wood-Vasey et al. (2008) and Friedman (2012). Aperture photometry with a 3'' aperture was performed at the position of SN 2009ip. Figure 2 presents the complete SN 2009ip NIR data set. The PAIRITEL photometry can be found in Table 6. A table will the UKIRT photometry will be published in M. Im et al. (in preparation).

In addition to the X-shooter spectra above, low-dispersion (R ≈ 700) NIR spectra were obtained with the 2.4 m Hiltner telescope at MDM Observatory on 2012 September 27 (t_pk − 6 days) and September 29 (t_pk − 4 days). The data were collected using TIFKAM. Data reduction followed standard procedures within IRAF. The spectra were corrected for telluric absorption and stellar features. Additional NIR low-resolution spectroscopy of SN 2009ip was obtained with the Folded-Port Infrared Echellette (FIRE) spectrograph (Simcoe et al. 2013) on the 6.5 m Magellan Baade Telescope on 2012 November 5, 25 and December 3. Data were reduced following Vacca et al. (2003), Foley et al. (2012) and Hsiao et al. (2013). Moderate-resolution (R ∼ 6000) NIR spectroscopy was obtained on 2012 November 19 with FIRE in high-resolution echellette mode. Data were reduced using a custom-developed IDL pipeline (FIREHOSE), evolved from the MASE suite used for optical echelle reduction (Bochanski et al. 2009). Standard procedures were followed to apply telluric corrections and relative flux calibrations as described above. The complete sequence of NIR spectra is shown in Figure 7. The observing log can be found in Table 3.

We obtained two sets of millimeter observations at mean frequency of ∼84.5 GHz (∼7.5 GHz bandwidth) with the Combined Array for Research in Millimeter Astronomy (CARMA; Bock et al. 2006) around maximum light, beginning on 2012 September 27.17 (t_pk − 6 days) and 2012 October 17.12 (t_pk + 14 days). We utilized 2158−150 and 2258−279 for gain calibration, 2232+117 for bandpass calibration, and Neptune for flux calibration. In ∼160 and ∼120 minute integration time at the position of SN 2009ip, we obtain 3σ upper limits on the flux density of 1.5 and 1.0 mJy, respectively. The overall flux uncertainty with CARMA is ∼20%. We observed the position of SN 2009ip with the Karl G. Jansky Very Large Array (VLA; Perley et al. 2011) on multiple epochs beginning on 2012 September 26.10 (t_pk − 7 days), with the last epoch beginning on 2012 December 2.93 (t_pk + 61 days). These observations were carried out at 21.25 GHz and 8.85 GHz with 2 GHz bandwidth in the VLA's most extended configuration (A; maximum baseline length ∼36.4 km) except for the first observations, which were obtained in the BnA configuration. In most epochs our observations of flux calibrator, 3C48, were too contaminated with radio frequency interference (RFI). Therefore, upon determining the flux of our gain calibrator J2213−2529 from the best observations of 3C48, we set the flux density of J2213−2529 in every epoch to be 0.65 and 0.63 Jy for 21.25 and 8.9 GHz, respectively. We note that this assumption might lead to slightly larger absolute flux uncertainties than usual (∼15%–20%). In addition, the source-phase calibrator cycle time (∼6 minutes) was a bit longer than standard for high frequency observations in an extended configuration, potentially increasing decoherence. We manually inspected the data and flagged edge channels and RFI, effectively reducing the bandwidth by ∼15%. We reduced all data using standard procedures in the Astronomical Image Processing System (Greisen 2003). A summary of the observations is presented in Table 7.

Swift-XRT data have been entirely acquired in photon counting mode and analyzed using the latest HEASOFT (v6.12) release, with standard filtering and screening criteria. For t < −11 days no X-ray source is detected at the position of SN 2009ip down to a 3σ limit of 3 × 10⁻³ cps in the 0.3–10 keV energy band (total exposure of 12.2 ks). In the time interval −11 days < t − t_pk < −2 days the 0.3–10 keV count-rate limit at the SN position is <1.1 × 10⁻³ cps (exposure time of 31.4 ks). Correcting for PSF losses and vignetting and merging the two time intervals we find no evidence for X-ray emission originating from SN 2009ip in the time interval −29 days < t − t_pk < −2 days down to a limit of <5.6 × 10⁻⁴ cps (0.3–10 keV, exposure time of 43.6 ks). An X-ray source is detected at the level of 5σ and 4σ in the time intervals −2 days < t − t_pk < +3 days and +4 days < t − t_pk < +13 days, respectively, with PSF and vignetting corrected count-rates of (1.6 ± 0.3) × 10⁻³ cps and (6.8 ± 1.8) × 10⁻⁴ cps (0.3–10 keV, exposure time of 42 and 44 ks).

We carried out XMM-Newton observations starting from 2012 November 3 (t_pk + 31 days). Observations have been obtained with the EPIC-PN and EPIC-MOS cameras in full frame with a thin filter mode. The total exposures for the EPIC-MOS1 and EPIC-MOS2 are 62.62 ks and 62.64 ks, respectively, and for the EPIC-PN, the exposure time is 54.82 ks. A point-like source is detected at the position of SN 2009ip with significance of 4.5σ (for EPIC-PN) and a rate of (2.7 ± 0.3) × 10⁻³ cps in a region of 10'' around the optical position of SN 2009ip. From 2013 January until April the source was Sun constrained for Swift. 10 ks of Swift-XRT data obtained on 2013 April 4.5 (t_pk + 183 days, when SN 2009ip became observable again) showed no detectable X-ray emission at the position of the transient down to a 3σ limit of 4.1 × 10⁻³ cps (0.3–10 keV).

An XMM EPIC-PN spectrum has been extracted with the SAS software selecting photons from a region of 10'' radius to avoid contamination from a nearby source (Figure 9). We model the spectrum with an absorption component (which combines the contribution from the Galaxy and from the SN 2009ip local environment, tbabs × ztbabs within Xspec) and an emission component. Both thermal bremsstrahlung and thermal emission from an optically thin plasma in collisional equilibrium (Xspec MEKAL model) can adequately fit the observed spectrum. In both cases we find kT > 10 keV and intrinsic hydrogen absorption of ${\rm NH_{{\rm int}}}\approx 10^{21} \,{\rm cm^{-2}}$. In the following we assume thermal bremsstrahlung emission with kT = 60 keV (this is the typical energy of photons expected from shock break-out from a dense CSM shell; see Section 7.5). We consider a nonthermal power-law emission model unlikely given the very hard best-fitting photon index of Γ = 0.87 ± 0.15 we obtain from this spectrum. The Galactic absorption in the direction of SN 2009ip is N_{H, MW} = 1.2 × 10²⁰ cm⁻² (Kalberla et al. 2005). The best-fitting neutral hydrogen intrinsic absorption is constrained to be ${\rm NH_{{\rm int}}}=0.10^{+0.06}_{-0.05}\times 10^{22}\ {\rm cm^{-2}}$. Using these parameters, the corresponding unabsorbed (absorbed) flux is (1.9 ± 0.2) × 10⁻¹⁴ erg s⁻¹ cm⁻² ((1.7 ± 0.2) × 10⁻¹⁴ erg s⁻¹ cm⁻²) in the 0.3–10 keV band. The spectrum is displayed in Figure 10. A Swift-XRT spectrum extracted around the peak (−2 days < t − t_pk < +13 days, total exposure of 86 ks) can be fit by a thermal bremsstrahlung model, assuming kT = 60 keV and ${\rm NH_{{\rm int}}}<3.1\times 10^{21}\ {\rm cm^{-2}}$ at the 3σ c.l. As for XMM, we use a 10'' extraction region to avoid contamination from a nearby source (Figure 9). The count-to-flux conversion factor deduced from this spectrum is $3.8\times 10^{-11}\ {\rm erg\,s^{-1}\,cm^{-2}\,{\rm count}^{-1}}$ (0.3–10 keV, unabsorbed). We use this factor to calibrate our Swift-XRT light-curve. The complete X-ray light-curve is shown in Figure 8.

We note that at the resolution of XMM and Swift-XRT we cannot exclude the presence of contaminating X-ray sources at a distance ≲ 10''. We further investigate this issue constraining the level of the contaminating flux by merging the Swift-XRT time intervals that yielded a nondetection at the SN 2009ip position. Using data collected between t_pk − 29 days and t_pk − 2 days, complemented by observations taken between t_pk + 29 days and t_pk + 90 days, we find evidence for an X-ray source located at R.A. = 22^h23^m0919 and decl. = −28°56'487 (J2000), with an uncertainty of 38 radius (90% containment), corresponding to 1'' from SN 2009ip. The source is detected at the level of 3.4σ with a PSF, vignetting, and exposure corrected count-rate of (3.0  ±  0.9) × 10⁻⁴ cps (total exposure of 110 ks, 0.3–10 keV energy band). The field is represented in Figure 9, left panel. This source contaminates the reported SN 2009ip flux at the level of ∼1.6 × 10⁻⁴ cps. Adopting the count-to-flux conversion factor above, this translates into a contaminating unabsorbed flux of ∼6 × 10⁻¹⁵ erg s⁻¹ cm⁻² (luminosity of ∼5 × 10³⁸ erg s⁻¹ at the distance of SN 2009ip), representing ∼10% the X-ray luminosity of SN 2009ip at peak. This source does not dominate the X-ray energy release around the peak time.

Swift-BAT data have been obtained in survey mode (15–150 keV energy range) between 2012 September 4 (t_pk − 29 days) and 2013 January 1 (t_pk + 90 days). We analyzed the data following standard procedures: sky images and source rates were obtained by processing the data with the batsurvey tool adopting standard screening and weighting based on the position of SN 2009ip. Fluxes were derived in the four standard energy channels, 14–24, 24–50, 50–100, and 100–195 keV. We converted the source rates to energy fluxes assuming a typical conversion factor of (5.9 ± 1.0) × 10⁻⁷ erg cm⁻² s⁻¹/count s⁻¹, estimated assuming a range of different photon indices of a power-law spectrum (Γ = 1–3). In particular, analyzing the data acquired around the optical peak we find evidence for a marginal detection at the level of 3.5σ in the time interval −0.8 days < t − t_pk < +0.2 days (corresponding to 2012 October 2.2–3.2). A spectrum extracted in this time interval can be fit by a power-law spectrum with photon index Γ = 1.8 ± 1.0 (90% c.l.), leading to a flux of (2.6 ± 1.4) × 10⁻¹⁰ erg s⁻¹ cm⁻² (90% c.l., 15–150 keV, and a total exposure time of 7.0 ks). We used this spectrum to flux calibrate the 5σ upper limit to the hard X-ray emission from SN 2009ip reported in Section 2.6 (F < 7 × 10⁻¹⁰ erg s⁻¹ cm⁻²).

Fermi-LAT observations have been obtained starting from 2012 September 3 (t_pk − 30 days) until 2012 October 31 (t_pk + 28 days). We note the presence of a data gap between t_pk − 9 days and t_pk − 2 days due to target-of-opportunity observations by Fermi during that time. We use events between 100 MeV and 10 GeV from the P7SOURCE_V6 data class (Ackermann et al. 2012), which is well suited for point-source analysis. Contamination from γ-rays produced by cosmic-ray interactions with the Earth's atmosphere is reduced by selecting events arriving at LAT within 100° of the zenith. Each interval is analyzed using a region of interest of 12° radius centered on the position of the source. In each time window, we performed a spectral analysis using the unbinned maximum likelihood algorithm gtlike. The background is modeled with two templates for diffuse γ-ray background emission: a Galactic component produced by the interaction of cosmic rays with the gas and interstellar radiation fields of the Milky Way, and an isotropic component that includes both the contribution of the extragalactic diffuse emission and the residual charged-particle backgrounds. The models used for this analysis, gal_2yearp7v6_v0.fits and iso_p7v6source.txt, are available from the Fermi Science Support Center, http://fermi.gsfc.nasa.gov/ssc/. This analysis uses the Fermi-LAT Science Tools, v. 09-28-00. We fix the normalization of the Galactic component but leave the normalization of the isotropic background as a free parameter. We also include the bright source 2FGL J2158.8−3013, located at approximately 548 from the location of SN 2009ip, and we fixed its parameters according to the values reported in Nolan et al. (2012).