CORE-COLLAPSE SUPERNOVAE MISSED BY OPTICAL SURVEYS

In order to investigate the completeness of the current local optical SN searches at different distances, we selected all the SNe discovered by the end of 2011 and after the beginning of the year 2000 from the Asiago SN catalog (Barbon et al. 1999) with an identified host galaxy and v_rec < 1500 km s⁻¹ (corresponding to distances less than ∼20 Mpc). We then adopted their radial velocities corrected for the peculiar velocities due to Virgo infall from the HyperLeda database (Paturel et al. 2003) and converted to distances. We ended up with a sample of 100 objects including also a few events without a spectroscopic classification, Type Ia's and luminous blue variable (LBV) outbursts originally classified as SNe. We then investigated the volumetric rate of these transients within different distances over the 12 yr period. A significantly higher rate was found at the smallest distances of ∼4–6 Mpc indicative of the local overdensity of star formation (see the discussion below). The rate was found to stay roughly constant between ∼12 and ∼15 Mpc and then show a significant decline. The drop of the rate at distances greater than ∼15 Mpc could be a result of the SN searches starting to miss a significant fraction of both the intrinsically faint and heavily dust-obscured events after this point (see also Botticella et al. 2012 and Horiuchi et al. 2011).

In the following analysis we therefore only consider the 49 events with Virgo infall corrected recession velocities of less than 1050 km s⁻¹ (or distances less than 15 Mpc). We excluded any Type Ia SNe (SNe 2001el, 2005df, 2005ew, 2006E, 2006mq, 2008ge, 2010ae, and 2011fe) and LBV outbursts (SNe 2000ch, 2002kg, and 2010da; Wagner et al. 2004; Pastorello et al. 2010; Smith et al. 2011; Maund et al. 2006). The initial sample also includes one event, SN 2008eh, without a spectroscopic classification. Based on its discovery magnitude and some light curve information, Horiuchi et al. (2011) assumed that it was an intrinsically faint CCSN. However, without any certain information on the nature of this event, we decide to exclude it from our final sample. Two of the remaining events with estimated distances larger than 15 Mpc were also excluded from our analysis. SN 2003hn occurred in NGC 1448, which also hosted the well-observed normal Type Ia SN 2001el. The optical and near-IR photometry of SN 2001el yielded a distance of 17.9 Mpc for NGC 1448 (Krisciunas et al. 2003; Mattila et al. 2005b). The host galaxy of SN 2004gk with a negative (Virgo infall corrected) recession velocity has a Tully–Fisher based distance of ∼20 Mpc (Solanes et al. 2002). We are therefore left with a total of 35 events which are listed in Table 1.

We also initially included SN 2008S-like events as a part of the analysis. We have therefore added the NGC 300-2008OT in Table 1, which is an event similar to SN 2008S and missing from the Asiago catalog. In total, our initial sample includes four such events: SNe 2002bu, 2008S, 2010dn, and NGC 300-2008OT. A number of authors (e.g., Botticella et al. 2009; Pumo et al. 2009) favor a scenario where the SN 2008 S-like events originate from an explosion of an extreme asymptotic giant branch star as an electron capture SN. However, several others found a non-explosive outburst of a massive star the most plausible origin (e.g., Bond et al. 2009; Berger et al. 2009; Smith et al. 2009; Kashi et al. 2010). Recent optical observations of the SN 2008S site (Szczygiel et al. 2012) have already ruled out the presence of a massive evolved progenitor star indicating that either it did not survive the 2008 event or has now returned to its dust enshrouded state. Kochanek (2011) predicts that the shock powering the current IR luminosity from the site of SN 2008S should destroy the dust, eventually allowing a direct confirmation if the progenitor of SN 2008S has disappeared. With the origin of the SN 2008S-like events still an open question, we decided to exclude them from our current analysis. We note that the SN 2008S-like events were also excluded by Botticella et al. (2012) from their CCSN rate analysis.

Our sample also includes the recently reported SN 2008jb that remained undetected by all the pointed SN searches despite a distance of only ∼9 Mpc (Prieto et al. 2012). It occurred in a low-luminosity host galaxy, ESO 302-G014, similar to the Magellanic Clouds and not included in the target lists of the pointed SN searches. The fraction of CCSNe occurring in such low-luminosity/metallicity galaxies is very low; <2% according to Young et al. (2008). Therefore, such events missed by optical-pointed SN searches locally are not likely to contribute significantly to the fraction of CCSNe missed by optical SN searches in general.

The adopted host galaxy distances are mostly the kinematic distances obtained from the Virgo infall corrected recession velocities. However, for several of the closest galaxies, more accurate distances were available from Cepheid observations or from the tip of the red giant branch (TRGB) method, which were adopted instead. In a few cases, the mean distance from several methods was adopted from the literature. The adopted distances are listed in Table 1 with the relevant references given in the notes to the table. In Table 1, we also list the IR luminosities of the host galaxies (if available from Sanders et al. 2003) scaled to our adopted distances.

The volumetric CCSN rates obtained using the events from Table 1 are listed in Table 2. We adopt the small number statistical uncertainties from Gehrels (1986) and use these throughout this study. Excluding the SN 2008S-like events there are a total of eight CCSNe within 6 Mpc, and the CCSN rate within 6 Mpc becomes 7.4^+3.7_{− 2.6} × 10⁻⁴ yr⁻¹ Mpc⁻³. This is significantly higher than the volumetric CCSN rate reported by Botticella et al. (2012), and that expected from the SFR extrapolated from high-z and even the locally normalized Horiuchi SFR. However, such an elevated CCSN rate is consistent with a local overdensity of star formation observed within ∼10 Mpc (e.g., Karachentsev et al. 2004). Furthermore, according to Heckman (1998), only four galaxies (M 82, NGC 253, M 83, and NGC 4945) are responsible for 25% of all the high-mass (⩾8 M_☉) star formation within 10 Mpc distance. It is interesting that all of these four galaxies are actually closer than 6 Mpc. Of the total of eight CCSNe within 6 Mpc, SNe 2004am and 2008iz occurred in M 82, SNe 2005af and 2011ja in NGC 4945, and SNe 2002hh and 2004et in NGC 6946. Thus, there are only five host galaxies within 6 Mpc that are responsible for all eight events over the last 12 yr within this volume. Also, it is possible that the SFR within 6 Mpc is actually concentrated in more dusty systems such as the prototypical starburst galaxy M 82, and, therefore, the missing fraction of SNe within this volume might be higher than representative for the local volume.

In order to avoid the effects of overdensity, we calculated the volumetric CCSN rates after excluding the volume within 6 Mpc. These rates are listed in Table 2 for different outer limits of the volume. We note that between 10 and 15 Mpc this volumetric CCSN rate stays constant within statistical uncertainties indicating that within 15 Mpc we are not missing a significant number of events. The volumetric CCSN rate at 6–15 Mpc is 1.5^+0.4_{− 0.3} × 10⁻⁴ yr⁻¹ Mpc⁻³, which is similar to the volumetric CCSN rate recently derived by Botticella et al. (2012) for a similar sample of CCSNe within 11 Mpc (the 14 events used by Botticella et al. are indicated in Column 11 of Table 1). This value is also consistent with the expected CCSN rate from the local star formation density (see Horiuchi et al. 2011).

For all the SNe within 12 Mpc except SN 2005at, host galaxy extinctions were available in the literature (for references see Column 7 of Table 1). We note that these extinctions were estimated for each SN, not for each galaxy as a whole. Therefore, we concentrate here on the CCSNe within 12 Mpc, which is almost identical to the distance limit in the sample of Botticella et al. (2012) when accounting for the difference in the value of H₀ adopted in the two studies. The Galactic extinctions were adopted from Schlegel et al. (1998). For SN 2005at, which is a Type Ic event resembling SN 1994I (see Schmidt & Salvo 2005) at ∼9 Mpc, very little information was available in the literature. In E. Kankare et al. (in preparation) we present the available UBVRI photometry of SN 2005at, together with the optical spectrum of Schmidt & Salvo (2005), covering a wide wavelength range from ∼3300 to ∼10200 Å. We use these data to estimate a host galaxy extinction of A_V = 2.3 ± 0.3 for SN 2005at by comparison with well-observed normal Type Ic events.

In Table 3, we compare the observed extinction properties of our CCSN sample within 12 Mpc with the predictions from our Monte Carlo simulations (E. Kankare et al. in preparation). For this comparison we include the 18 CCSNe from Table 1, excluding the SN 2008S-like events whose origin is still an open question. These simulations follow the recipe in Riello & Patat (2005), assuming a homogeneous dust distribution for the model host galaxy with the bulge-to-total (B/T) ratio = 0.0 (100% of the SNe located in the disk) and an optical depth through a simulated face-on galaxy at a zero radius of τ_V(0) = 2.5, consistent with many statistical studies (e.g., Kankare et al. 2009), to derive a distribution of expected A_V values. For the purpose of comparing observed and predicted extinction values, we divided our sample into three inclination bins. The host galaxy inclinations were adopted from the HyperLeda database.

First, we compare the observed fraction of CCSNe in different inclination bins with the predictions from the simulations (see Columns 7 and 11 in Table 3). The relative distribution of SNe in galaxies with inclinations of 0°–30° and 30°–60° appears to be as expected, whereas there is an apparent lack of SNe in host galaxies with inclinations higher than 60°. Although half of the SNe would be expected in galaxies with an inclination of 61°–90°, less than 30% were discovered in such galaxies. In the edge-on bin, ∼10% of the SNe have a simulated extinction of A_V > 5. Therefore, we do expect a fraction of the SNe to be missed in normal galaxies with the highest inclinations. However, the expected median extinction of objects in the edge-on bin is ∼0.6 in A_V, which means that the optical SN searches should have been sensitive to most of the SNe even in the edge-on galaxies within the 12 Mpc volume. Therefore, the apparent lack of events in the edge-on bin appears to be more likely due to the selection effect of the SN searches avoiding these galaxies rather than being a consequence of the host galaxy extinction. The fact that there is an apparent lack of CCSNe with edge-on host galaxies indicates that our SN sample within 12 Mpc (and the one used by Botticella et al. 2012) is likely not complete. Therefore, the local volumetric CCSN rate could be higher than estimated above, which indicates that the local overdensity of star formation could also affect the results within the 6–15 Mpc distance.

Second, we compare the observed host galaxy extinctions in different inclination bins with the predictions (see Table 3). In the face-on (0°–30°) bin the observed average and median extinctions of A_V = 0.18 and 0.15, respectively, are quite low, similar to the predicted values, although any quantitative comparison between the extinctions is difficult because of the small number of events. However, in the intermediate inclination (30°–60°) bin we have 10 SNe, allowing for a more meaningful comparison. The average and median extinctions for this sample are 1.25 and 0.40 in A_V, which are both significantly higher than their predicted values of 0.52 and 0.20, respectively, indicating that the smooth distribution of dust used in the simulations does not agree well with the observations. In the edge-on (60°–90°) sample we have five events, two of them (SNe 2004am and 2008iz) within the nuclear regions of the prototypical nearby starburst galaxy M 82. SN 2008iz was discovered at radio wavelengths with no reported detection in the optical indicating a very high extinction being within the nuclear regions of M 82 (e.g., Brunthaler et al. 2010). Based on a detection in the near-IR K-band, Mattila et al. (2012) estimated the likely extinction towards SN 2008iz to be not more than A_V ∼ 10. SN 2004am occurred coincident with the obscured super star cluster M82-L and has an estimated host galaxy extinction of A_V ∼ 5 (Mattila et al. 2012). Excluding these two events, the average extinction for the remaining three events occurring in normal spiral galaxies is ∼1.1, which is lower than predicted but of course highly uncertain due to the very small number of events included.

Using our adopted distances and extinctions we calculated optical absolute peak magnitudes for the events within a distance of 12 Mpc. These are mostly based on the SN peak magnitudes available from Horiuchi et al. (2011) and Botticella et al. (2012). These absolute SN peak magnitudes are very approximate, especially for a couple of the most recent events for which we adopted their discovery magnitudes (for references see the note to Table 1). We find that 3 of the 17 CCSNe within 12 Mpc are fainter than M = −15.5, but none of them are fainter than M = −15.0. We note that the absolute SN magnitudes considered by Horiuchi et al. (2011) were not corrected for host galaxy extinctions. After applying the correction, we find from their intrinsically faint CCSNe that only SN 2004am has an absolute peak magnitude of ∼ −15 (see Column 9 in Table 1). These faint SNe are more likely to be missed in a search over a larger volume compared to the restriction <12 Mpc for our sample and could therefore lead to underestimates of the CCSN rate. The absolute peak magnitudes listed in Table 1 (excluding SN 2008iz with no reported optical detection) therefore suggest that a correction factor approximately ∼1.2 may actually be needed to account for the intrinsically faint CCSNe in a survey that is not deep enough for their detection. However, such intrinsically faint events are best accounted for by including a realistic distribution of the absolute peak magnitudes when deriving the CCSN rate results. For example, the effects of including a large (30%) fraction of faint (M > −15) CCSNe was studied by Melinder et al. (2012). As a result, they found an increase of ∼30% in their CCSN rates at z ∼ 0.4 and z ∼ 0.7.

We can now proceed to estimate the missing fraction of SNe due to high dust extinctions in normal galaxies (L_IR < 10¹¹ L_☉) using all the available information we have gathered. A correction for the dust extinction along the lines of Riello & Patat (2005) can be used to compensate for the CCSNe missed due to inclination effects. However, this should be combined with a fraction of SNe with significantly higher extinctions than predicted by these models when considering a realistic distribution of extinctions for CCSNe in normal galaxies. We concentrate on the sample of 13 CCSNe with host galaxy inclinations of 0°–60° (see Table 3). The host galaxies of these events have IR luminosities ranging between ∼10⁹ and ∼3 × 10¹⁰L_☉ (see Column 12, Table 1). The observed average extinction calculated for the 0°–60° bin of 〈A_V〉 = 1.00 (sample 1 in Table 3) is significantly higher than the predicted value, 〈A_V〉 = 0.48 (although the median extinctions are similar). However, if excluding the two events with the highest host galaxy extinctions (SNe 2002hh and 2009hd with A_V = 3.7 and 4.1, respectively), the observed average extinction becomes 〈A_V〉 = 0.48 (sample 2 in Table 3), which is identical to the predicted value (the median value does not change). The standard deviations of the observed and predicted distributions are also now very similar (0.74 versus 0.68).

In Figure 1, the observed and predicted extinction distributions are compared after excluding the two outliers with A_V ≳ 3.7. We note that only about 0.3% of the events in the predicted distribution suffer from extinctions higher than A_V = 3.7. This indicates that the simple model with a homogeneous dust distribution used for the simulations does not produce a realistic distribution of line-of-sight extinctions for SNe in spiral galaxies. CCSNe are observed to be associated with star-forming regions in their host galaxies (e.g., Anderson et al. 2012) and therefore higher extinction values local to the SNe can be expected. However, we also note that the number of SNe in our current sample is rather small, making any detailed comparison less reliable.

Zoom In Zoom Out Reset image size

The two outlier events with host galaxy extinctions of A_V = 3.7–4.1 (or A_B = 4.9–5.5) would be the ones most likely missed by optical SN searches in more distant galaxies. For example, a typical Type II-P event with absolute peak magnitude M(B) = −17.0 at a redshift of 0.5 would have an apparent peak magnitude of m(R) ∼ 30 if the host galaxy extinction is A_B = 5. This is several magnitudes below the limits of the current high-z SN searches (e.g., Dahlen et al. 2012; Melinder et al. 2012). At higher redshifts the effects of the extinction would be even more severe for an optical SN search since shorter rest-frame wavelengths are observed. However, we note that SN searches are often optimized for detecting SNe at rest-frame B and V bands, since the SN spectral energy distribution (SED) peaks in this range. For example, searches aiming at a redshift around 0.5 (e.g., Bazin et al. 2009; Melinder et al. 2011) have been typically observing in R and I bands and searches reaching for z ∼ 1 (e.g., Dahlen et al. 2012) in the z-band. The ongoing SN search program (e.g., Rodney et al. 2012) as a part of the CANDELS project (e.g., Grogin et al. 2011) is aiming to detect SNe at z ∼1.5–2 and observes in the near-IR J and H bands. Therefore, we now use the sample of 13 CCSNe within the 0°–60° bin and the two outlier events with the highest host galaxy extinctions to estimate the fraction of CCSNe likely missed by rest-frame optical surveys in normal galaxies to be 15⁺²¹_{− 10}%.

During the last two decades, Arp 299 has been the target of a number of professional and amateur SN searches (see Table 4 for summary). At optical wavelengths, Arp 299 has been the target of at least four professional SN searches. In the early 1990s it was monitored by the Leuschner Observatory SN Search that was the predecessor of the Lick Observatory SN search (LOSS), the Beijing Observatory SN search, the Richmond et al. (1998) search, and finally the LOSS. Arp 299 was also included in the "optimal" galaxy sample of the LOSS with their SN search data between 1998 March and 2008 December used for the SN rate calculation (Leaman et al. 2011; Li et al. 2011a, 2011b). SN 1993G was discovered by the Leuschner Observatory SN Search, SN 1998T by the Beijing Observatory SN search, and SN 1999D by the LOSS. LOSS also detected SNe 1998T and 2005U, although they were originally discovered elsewhere. The observed B − V color of SN 1993G indicated that the SN was virtually unreddened (Tsvetkov 1994). SN 1998T was spectroscopically classified by Li et al. (1998) reporting strong P-Cygni profiles of He i lines between 510 and 900 nm. Therefore, the extinction toward SN 1998T was also likely low. In the case of SN 1999D the typing spectrum featured a very blue continuum (Jha et al. 1999), strongly suggesting a low extinction. For these three events we therefore assume that the host galaxy extinction was close to zero.

Van Buren et al. (1994) conducted a K-band survey for SNe in starburst galaxies at the NASA 3.0 m Infrared Telescope Facility (IRTF). As a part of their search Arp 299 was observed in at least four epochs between 1992 March and 1993 December, yielding the discovery of SN 1992bu in their K-band images. The SN is located at a projected distance of ∼1.0 kpc from the nucleus B1 of the galaxy. SN 1992bu has no spectroscopic classification and, based on just the K-band photometry, estimating the extinction is difficult. However, as a part of their analysis, Anderson et al. (2011) noted that SN 1992bu falls on a bright star-forming region and has a small galactocentric distance. Hence, they find it consistent with being a stripped envelope core-collapse event.

A more recent near-IR SN search targeting nearby starburst galaxies was carried out in the K' band by Grossan et al. (1999). The observations were performed with the 2.3 m telescope at the Wyoming IR Observatory (WIRO). As part of the search, five observations of Arp 299 were taken between 1993 February and December. Arp 299 was also targeted for a near-IR K-band SN search carried out with the 3.6 m Telescopio Nazionale Galileo (TNG) between 1999 October and 2001 October including nine observations (Mannucci et al. 2003). Another near-IR K-band SN search observing Arp 299 was carried out using the 1.55 m U.S. Naval Observatory telescope at Flagstaff (e.g., Dudley et al. 2008). During the search, Arp 299 was observed in a total of 19 epochs between 2001 February and 2004 May (C. C. Dudley 2012, private communication). However, no confirmed SNe were reported in Arp 299 as a result of these three searches.

Arp 299 was also monitored in a near-IR K-band for a total of eight epochs between 2002 January and 2005 January as a part of the nuclear SN search campaign with the William Herschel Telescope resulting in the discovery of SN 2005U (Mattila et al. 2004, 2005a). The near-IR color of SN 2005U indicated only a modest extinction. The SN was spectroscopically classified as a Type-II event, probably within a few weeks of the explosion (Modjaz et al. 2005). A further spectrum of SN 2005U was obtained by Leonard & Cenko (2005) showing it to resemble the spectrum of the Type IIb SN 1993J over a month past its explosion with no evidence of substantial host galaxy extinction (D. C. Leonard 2011, private communication; M. Modjaz et al., in preparation).

More recently, we have been monitoring Arp 299 for SNe in the K band using the 2.6 m Nordic Optical Telescope (NOT) every few months for a total of ∼20 epochs, resulting in the discovery of SN 2010P (Mattila & Kankare 2010) and the detection of SN 2010O, which was also discovered at optical wavelengths (Newton et al. 2010). SN 2010O was spectroscopically classified as a Type Ib/c event around maximum light and a host galaxy extinction of A_V ∼ 1.9 was estimated from prominent Na i D absorption lines from the host galaxy (Mattila et al. 2010). Between 2008 and 2011 we have also been monitoring Arp 299 using high spatial resolution observations with laser guide star adaptive optics on the Gemini North telescope as part of a sample of nearby LIRGs (see Kankare et al. 2008, 2012). Our most recent K-band observations from the NOT and the Gemini-N telescope were obtained in 2012 March and 2012 February, respectively.

Recently, Anderson et al. (2011) analyzed the circumnuclear SN population in Arp 299 finding a relatively high fraction of stripped envelope events (Types Ib and IIb) relative to other Type II SNe. They suggested that this excess could be explained by the young age of circumnuclear star formation in Arp 299 such that we would now be witnessing the explosions of the most massive stars formed. Alternatively, the excess of stripped envelope events might be explained by a top-heavy initial mass function (IMF) favoring the formation of the most massive stars.

3.1.1. The Highly Obscured SN 2010P

SN 2010P was discovered via image-subtraction techniques in near-IR images obtained using NOTCam on January 18.2 and 23.1 UT. The approximate magnitudes of SN 2010P on 2010 January 23.1 were m_I = 18.3, m_J = 16.8, m_H = 16.2, and m_K = 15.9 (Mattila & Kankare 2010). Nothing is visible at this position on NOTCam K-band images taken on 2009 November 27.2 (limiting mag 17.5). We note that SN 2010P was also detectable by image subtraction in our I-band image obtained using StanCam on January 23.1 on the NOT. However, the SN was not detected in our R-band images from the same instrument. A radio follow-up using MERLIN between 2010 January 29 and February 1 at 4.99 GHz resulted in a non-detection for SN 2010P (Beswick et al. 2010). The absolute magnitude and colors of 2010P are consistent with the CCSN template light curves from Mattila & Meikle (2001) with a likely extinction of A_V ∼ 5.

In order to allow spectroscopic typing and a more reliable extinction determination, we obtained long-slit spectroscopy of SN 2010P with the Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) attached to the Gemini-N Telescope as a part of the programGN-2010A-Q-40 (PI: S. Ryder). Four exposures of 900 s each were taken on the night of 2010 February 11 UT using the R400 grating and a 05 wide slit to deliver a resolution of R ∼ 1900. The data were reduced and combined using V1.10 of the gemini package within IRAF.⁹

The spectrum of SN 2010P has a red continuum and very little signal at wavelengths shorter than 6000 Å (see Figure 3), indicative of a high extinction. Despite significant reddening, the spectrum was found to be consistent with an H-deficient CCSN. Comparison with a library of SN spectra using both the "GELATO" code (Harutyunyan et al. 2008) and the "SuperNova IDentification" code (Blondin & Tonry 2007) yielded good matches to several Type Ib and Type IIb SNe between one and three weeks after maximum light (the spectrum of SN 2010P was dereddened by 5 mag of extinction in A_V before using the "Gelato" code). In Figure 3, we illustrate this by comparing the spectrum of SN 2010P, dereddened by A_V = 3, 5, and 7, with that of the Type IIb/Ib event SN 2000H at 5 days past maximum from Branch et al. (2002). For this the reddening law of Cardelli et al. (1989) was assumed with a value of R_V = 3.1. Following Elmhamdi et al. (2006) we assume the host galaxy extinction toward SN 2000H to be small. A very high amount of extinction toward SN 2010P is also consistent with its near-IR light curves, colors, and absolute magnitude.

Zoom In Zoom Out Reset image size

SN 2010P has a projected distance of only ∼200 pc from the source C', but ∼1.3 kpc from the source C (see Figure 2). Although the region C' is a strong radio source, it is much less prominent at IR wavelengths with only ∼10% of source C at 2.2 μm increasing to ∼1/3 at 18 μm (Charmandaris et al. 2002). Based on the optical depth of the 9.7 μm absorption feature, Alonso-Herrero et al. (2009) estimated an extinction of A_V = 12 toward source C', which is similar to their estimated extinction for nucleus B1. Having only ∼200 pc projected distance from the source C' and a substantial host galaxy extinction, we consider SN 2010P as a genuine nuclear SN in the following analysis. Other SNe with high host galaxy extinctions previously detected in LIRGs include SNe 2001db (A_V ∼ 5.5; Maiolino et al. 2002), 2004ip (A_V ⩾ 5; Mattila et al. 2007), 2008cs (A_V ∼ 16; Kankare et al. 2008), and 2011hi (A_V = 5–7; Kankare et al. 2012; Romero-Cañizales et al. 2012), all discovered in the near-IR K band similar to SN 2010P.

Including only the SNe discovered from 1998 onward when the LOSS and a number of near-IR SN searches monitoring Arp 299 were active, we have a total of five SNe discovered during a period of ∼14 yr (1998 March–2012 March). The projected distances of these SNe from the nearest nuclei are listed in Table 5. All of these except for SN 2010P range between ∼1 and ∼5 kpc.

Despite the numerous efforts at optical and near-IR wavelengths to detect SNe within the innermost nuclear regions of Arp 299 during the last two decades, only one (SN 2010P) has been detected close to one of the nuclei of Arp 299. This is consistent with the high dust extinctions therein, which make the SNe too dim to be detectable. In order to estimate the actual fraction of SNe missed in Arp 299 due to dust extinction, we need to estimate the fraction of star formation and thus CCSNe in the nuclear and circumnuclear regions.

From the observations we know that at least five SNe occurred in Arp 299 during the 14 yr period. Using these five SNe we can now estimate a lower limit for the optical/near-IR CCSN rate in Arp 299 over the period of 14 yr of 0.36^+0.24_{− 0.15} yr⁻¹. Of the five SNe all except SN 2010P were detected by the optical SN searches (although SN 2005U was originally discovered in the near-IR) and had a relatively low extinction. SN 2010P was marginally detected in our I-band images from the NOT (after its initial discovery in the near-IR) and was not detectable in R-band images from the same telescope. It had a high extinction of A_V ∼ 5 and therefore most likely would not be detected in the current optical SN searches. In fact, shortly before our discovery report on SN 2010P, the discovery of SN 2010O was reported by Newton et al. (2010) in the course of the Puckett Observatory Supernova Search. They did not detect SN 2010P in their unfiltered CCD images. Therefore, we assume that 4 out of the 5 SNe discovered in Arp 299 during the 14 yr period were optically detectable. This yields a lower limit for the rate of optically detectable circumnuclear SNe in Arp 299 of 0.29^+0.23_{− 0.14} yr⁻¹.

After having determined the observed SN rate in Arp 299, we now turn to estimating the predicted rate based on the SFR in order to calculate the fraction of SNe hidden by dust. The CCSN rate within the highly obscured innermost nuclear regions of Arp 299 can be estimated either via infrared or radio luminosity. Radio luminosity based estimates for the nuclear SN rate of Arp 299 from VLA measurements have been presented by Neff et al. (2004) who estimated (scaled to the distance of 46.7 Mpc assumed here) 0.65–1.3 yr⁻¹, 0.13–0.26 yr⁻¹, and 0.07–0.13 and 0.04–0.08 yr⁻¹ for the nuclei A and B1, and sources C and C', respectively (see Figure 2 for the identifications of the different nuclei). Alonso-Herrero et al. (2000) estimated 0.8 yr⁻¹ and 0.15 yr⁻¹ for nuclei A and B, respectively (again scaled to 46.7 Mpc). More, recently, Romero-Cañizales et al. (2011) estimated a lower limit of 0.28^+0.27_{− 0.15} yr⁻¹ for the CCSN rate of nucleus B1 based on archival VLA observations over a period of 11 yr, which we will adopt as the more direct estimate to use in this study. Pérez-Torres et al. (2009) and Bondi et al. (2012) made use of extremely high spatial resolution European VLBI network (EVN) radio observations of the innermost nuclear regions of Arp 299-A. They find clear evidence for at least two new radio supernovae (RSNe) in their observations separated by two years therefore implying a lower limit for the CCSN rate in nucleus A of ∼0.8 yr⁻¹.

We can also estimate the CCSN rates indirectly from the galaxy IR luminosity. Charmandaris et al. (2002) estimated that IC 694 (nucleus A), NGC 3690 (nucleus B1+B2), and sources C+C' emit approximately 39%, 20%, and 10% of the total IR luminosity of Arp 299, respectively, with the remaining ∼31% originating from the circumnuclear regions (see Column 2, Table 7). The rather large contribution of the circumnuclear regions to the total luminosity is also supported by the Hα observations by García-Marín et al. (2006) showing significant amounts of star formation in the spiral arms of IC 694 and also to the west of nucleus B1 in NGC 3690. Adopting the galaxy IR luminosity of 7.3 × 10¹¹ L_☉ and an empirical relation between the IR luminosity and CCSN rate (r_SN = 2.7 × 10⁻¹² × L_IR/L_☉ yr⁻¹) from Mattila & Meikle (2001), we have a total CCSN rate of ∼2.0 yr⁻¹ for the entire system (see also the discussion in Romero-Cañizales et al. 2011). However, this IR-luminosity-based CCSN rate estimate could have a significant uncertainty because the empirical relation is based on only three nearby starburst galaxies NGC 253, M 82, and NGC 4038/9 with IR luminosities substantially lower than of Arp 299.

As a more accurate approach we estimate the CCSN rates for the Arp 299 nuclei by modeling their SEDs. This approach has the advantage that it can take into account the effects of the starburst age as well as the possible contribution of an active galactic nucleus (AGN) to the IR luminosities of the nuclei. For this purpose we have used low-resolution mid-IR Spitzer/IRS (SL+LL setting; 5–38 μm range) spectra covering ∼104 × 104 rectangular regions (see the lower panel of Figure 5 in Alonso-Herrero et al. 2009) centered on the nuclei A and B1+B2. These spectra were obtained on 2004 April 15 and have already been reported in Alonso-Herrero et al. (2009). In addition, we included IRAS 12, 25, 60, and 100 μm fluxes from Sanders et al. (2003) assigned to nuclei A and B1+B2 according to the estimated contributions from Charmandaris et al. (2002). In order to obtain a reasonable match between the Spitzer spectra and the IRAS 12 and 25 μm fluxes, we multiplied both the Spitzer/IRS spectra by 1.3. This value is well within the uncertainties expected in the estimates of Charmandaris et al. (2002) for the fractions of IR luminosities arising from the different nuclei of Arp 299 and the calibration errors of the IRAS and Spitzer/IRS data.

For modeling the SED of Arp 299 we use a grid of AGN torus models that have been computed with the method of Efstathiou & Rowan-Robinson (1995) and a grid of starburst models that have been computed with the method of Efstathiou et al. (2000). For the AGN torus models we use the tapered disk models computed with the method of Efstathiou & Rowan-Robinson (1995) and described in more detail in A. Efstathiou et al. (2012, in preparation). These models considered a distribution of grain species and sizes, multiple scattering and a density distribution that followed r⁻¹ where r is the distance from the central source. The models assumed a smooth distribution of dust, so they are a good approximation of the density distribution in the torus if the mean distance between clouds is small compared with the size of the torus, but they have been quite successful in fitting the SEDs of AGNs even in cases where mid-infrared spectroscopy is available (e.g., A. Efstathiou et al. 2012, in preparation). In this grid of models we consider four discrete values for the equatorial 1000 Å optical depth (500, 750, 1000, 1250), three values for the ratio of outer-to-inner disk radii (20, 60, 100) and three values for the opening angle of the disk (30°, 45°, and 60°). The spectra are computed for inclinations which are equally spaced in the range 0–π/2.

Efstathiou et al. (2000) presented a starburst model that combined the stellar population synthesis model of Bruzual & Charlot (2003), a detailed radiative transfer that included the effect of small grains and polycyclic aromatic hydrocarbons (PAHs), and a simple evolutionary scheme for the molecular clouds that constitute the starburst. The model predicts the SEDs of starburst galaxies from the ultraviolet to the millimeter as a function of the age of the starburst and the initial optical depth of the molecular clouds. In this paper, we use a sequence of models that have been computed with an updated dust model (Efstathiou & Siebenmorgen 2009). We assume an exponentially declining SFR with an e-folding time of 20 Myr. The choice of 20 Myr is supported by fits to the far-IR color–color diagrams (e.g., Efstathiou et al. 2000), to the SEDs of ULIRGs (e.g., Farrah et al. 2003), and to the Spoon diagram (Rowan-Robinson & Efstathiou 2009). Our model fits to the Spitzer/IRS (SL+LL) spectra are shown in Figure 4. For nucleus A, no AGN component is required, whereas for nucleus B1+B2 we find a 20% contribution by an AGN to the total IR (8–1000 μm) luminosity. This agrees well with the recent findings of Alonso-Herrero et al. (2012). The best-fit dusty torus parameters we find for the AGN are τ(1000 Å) = 500, r_outer/r_inner = 20, an opening angle of 60°, and an inclination of 45°. The resulting starburst parameters are listed in Table 6.

Zoom In Zoom Out Reset image size

The CCSN rates were then estimated in the following way. The stellar population synthesis model of Bruzual & Charlot (2003) makes a prediction of the CCSN rate SNR(t) at a time t after star formation in an instantaneous burst. The starburst model of Efstathiou et al. (2000) predicts the spectrum of this instantaneous burst at time t and assumes a star formation history for the starburst. It is therefore possible to calculate self-consistently the CCSN rate at different stages in the evolution of a starburst by convolving the star formation history with SNR(t). This results in CCSN rates of 0.76 and 0.33 yr⁻¹ for nuclei A and B1+B2, respectively.

The IR luminosities for the different components of Arp 299 are listed in Table 7 (Column 2) adopting their fractional contributions from Charmandaris et al. (2002) and a total IR luminosity for the system of 7.3 × 10¹¹ L_☉. We compared our SED fit based model IR luminosities (see Table 6) of nuclei A and B1+B2 with the "observed" ones obtained as described above. The model luminosity for nucleus A of 2.45 × 10¹¹ L_☉ is slightly lower than the observed L_IR = 2.85 × 10¹¹ L_☉. However, we note that the simple gray body models of Charmandaris et al. (2002) also gave lower IR luminosities for the Arp 299 nuclei than the observed values. The total (AGN+starburst) model IR luminosity for nucleus B1+B2 of 1.49 × 10¹¹ L_☉ is almost identical to the observed value of 1.46 × 10¹¹ L_☉.

For the sources C+C' we estimate the CCSN rates from the observed IR luminosity using the empirical relation of Mattila & Meikle (2001). This yields a CCSN rate of 0.20 yr⁻¹ which is also consistent with the radio-based estimate of 0.16 ± 0.05 yr⁻¹. We note that the same approach would yield CCSN rate estimates of 0.77 and 0.39 yr⁻¹ for nuclei A and B1+B2 very similar to the IR SED modeling based values (note that a slightly higher value for nucleus B1+B2 is as expected due to the 20% AGN contribution in B1). For the total nuclear CCSN rate we combine the results from the IR SED modeling of nucleus A and B1+B2, with the IR luminosity derived rate for sources C+C' to obtain a predicted nuclear CCSN rate of 1.29 yr⁻¹.

The results from Charmandaris et al. (2002) suggest that 31% of the total IR luminosity originates outside the nuclei A, B1+B2, and C+C'. Furthermore, Alonso-Herrero et al. (2009) also found evidence from their Spitzer/IRS spectral mapping for significant circumnuclear PAH and [Ne ii] emission extending outside the main nuclei in Arp 299. Adopting the empirical relation between the IR luminosity and the CCSN rate, this corresponds to 0.61 yr⁻¹ for the circumnuclear regions. We note that the use of the CCSN rate estimate (for unobscured Type II+Ib/c SNe in normal galaxies) in units of the galaxy far-IR luminosity from Cappellaro et al. (1999) would yield a very similar circumnuclear CCSN rate estimate. Furthermore, García-Marín et al. (2006) determined an SFR of 43 M_☉ yr⁻¹ for the entire Arp 299 system (with the nucleus B1 excluded) based on their estimate for the extinction-corrected Hα luminosity. García-Marín et al. (2006) found that most of the circumnuclear H ii regions in Arp 299 only suffer from modest extinctions of typically less than A_V ∼ 1. Assuming a Salpeter IMF between 0.1 and 125 M_☉ and CCSN progenitor masses between 8 and 50 M_☉ (e.g., see the discussion in Melinder et al. 2012 and Dahlen et al. 2012), this corresponds to a CCSN rate of ∼0.30 yr⁻¹ for Arp 299. Having the Hα emission from Arp 299 arising mostly outside the heavily obscured nuclear regions A, B1+B2, and C+C' (in contrast to the IR luminosity), we can consider this as a robust lower limit for the circumnuclear CCSN rate and adopt the IR-luminosity-based value of 0.61 yr⁻¹ as an upper limit. Combining these circumnuclear CCSN rate estimates with the nuclear CCSN rate therefore yields a total CCSN rate of 1.59–1.90 yr⁻¹ for Arp 299. In Table 7, our predicted CCSN rates are compared to the observed rates based on radio, optical+NIR, and optical searches. These are broadly consistent with each other and together with the data on the optical SN discoveries in the circumnuclear regions can be used to estimate the missing fraction of SNe in Arp 299.

We are now ready to estimate the missing fractions. Within the nuclear regions, no SNe have been detected by optical observations (and only one, SN 2010P, by near-IR observations) and we therefore assume that 100% of the SNe are missed by optical searches. For the circumnuclear regions, we estimated a lower limit for the "optical" CCSN rate of 0.29^+0.23_{− 0.14} yr⁻¹, while the predicted rate is 0.30–0.61 yr⁻¹, suggesting a missing fraction of up to 37⁺³⁸_{− 37} % in this region. Compared to the total predicted CCSN rate in Arp 299 of 1.59–1.91 yr⁻¹, we estimate a missing fraction of up to 83⁺⁹_{− 15} %, taking into account both statistical errors and the uncertainty in the total CCSN rate. We adopt these values as the missing fractions of CCSNe in local LIRGs (see Table 8).