Astronomical Science

Search for Supernovae in Starburst Galaxies with HAWK-I

Matteo Miluzio1 the whole history of , due to An infrared search in starburst galaxies: the wide range in the delay times of their Why? progenitors. Recently, it was claimed that 1 INAF–Osservatorio Astronomico di a significant fraction of SN Ia have a short Starburst galaxies have very high SFRs, 7 Padova, Italy delay time of about 10 years (Mannucci in the range of 10–100 MA/year, com- et al, 2006). As with CC SN, the rate of pared to the few solar masses per year such prompt SNe Ia events is proportional of normal­ star-forming galaxies. Many With the aim of testing the relation to the current SFR. starburst galaxies are in close pairs or between (SN) rate and star have disturbed morphology, which is a formation rate, we conducted a SN sign that merging is enhancing their SFR. search in a sample of local starburst Motivations: The SN rate problem Since the ultraviolet radiation from young galaxies (SBs) where both star forma- and massive stars heats the surrounding tion rates and extinction are extremely During the last decade significant effort dust and is re-emitted in the far-infrared, high. The search was performed in the has been devoted to the measurement of the most luminous starbursts are Lumi- near-infrared, where the bias due to the cosmic SFR, using a combination nous Infrared Galaxies (LIRGS) with 1011 12 extinction is reduced using HAWK-I on of many different probes (e.g., Hopkins & < LIR <10 LA and UltraLuminous Infrared 12 the VLT. We discovered six SNe, in Beacom, 2006). Also, as a result of a Galaxies (ULIRGs) with LIR > 10 LA. excellent agreement with expectations, number of dedicated surveys, new esti- when considering that, even in our mates of the SN rate have become avail­ The first attempts to measure SN rates in search, about 60% of events remain able. Although the focus was on type starburst galaxies in the optical band date hidden in the nuclear regions due to a Ia SN, a few estimates of the CC SN rates back to 1999 (Richmond et al., 1998) with combination of reduced search effi- have been published (e.g., Melinder et al., only a handful of events detected. The ciency and very high extinction. 2012). In particular, the evolution of SN first results of a systematic search in star- rates with was found to track the burst galaxies in the infrared (IR) domain SFR evolution very well, considering the were reported by Maiolino et al. (2002) and Why supernova rates? large uncertainties in the extinction cor- Mannucci et al. (2003). The main conclu- rections. Botticella et al. (2008) obtained a sion was that the observed SN rate in star- The determination of the star formation good match between the observed SN burst galaxies is one order of magnitude history (SFH) of the Universe is one of and star formation rates assuming a lower higher than that expected from the blue the main goals of modern cosmology, as limit for CC SN progenitors of 10 MA. luminosity of the galaxies, but still three to it is crucial to our understanding of how ten times lower than would be ex­­pected structures in the Universe form and However, recent study of the core col- from the far-IR luminosity. They suggested evolve. The estimate of the star formation lapse SN progenitors, based on direct that the most plausible explanation for this rate mainly relies on measurements of detection on pre-explosion archival discrepancy was the extreme extinction in integrated galaxy luminosity, most often images, set a lower limit of 8 MA (Smartt the galaxy nuclear regions (AV > 25 mag) in the infrared or for specific emission et al., 2009). It was claimed that if this that hides SNe even in the IR. Recently, lines, e.g., Hα. The luminosity measure- value is adopted, the observed SN rates Mattila et al. (2007) conducted an IR ments are converted to total luminosity would be a factor of two lower than those search in starburst galaxies using ground- assuming an initial mass function (IMF) expected from the observed SFR, some- based telescopes with adaptive optics, a for the stellar population and including a times called the SN rate problem (e.g., technique that allows the nuclear regions number of corrections to account for Horiuchi et al., 2011). While we should to be observed with high spatial resolution. the presence of multiple stellar popula- keep in mind the uncertainties in SFR cal- This approach led to the discovery of a tions, non-thermal source contributions ibrations, there are also possible impor- handful of SNe, but not yet an estimate of and extinction. tant biases in the SN rate estimates. The the CC SN rate. About a dozen SNe have two most severe are the possible exist- been discovered by IR searches so far. In principle a direct test of the star forma- ence of a large population of very faint Therefore the statistics are still based on tion rate (SFR) estimate can be achieved CC SNe and/or underestimates of the very low numbers, and many of the original by directly counting the number of dying correction for extinction. In particular, questions are still unanswered. Entering massive stars, namely core collapse several authors (e.g., Mattila et al., 2012) into this debate, we started a new search supernovae (CC SNe). Indeed, because argue that up to 70–90% of CC SNe to measure the SN rate using the High of their short-lived progenitors, CC SNe remain hidden in the highly dusty nuclear Acuity Wide-fieldK -band Imager (HAWK-I), trace the current SFR. On the other hand, regions of starburst galaxies. If a correc- the infrared camera mounted at the ESO for an adopted SFR, measurements of tion for the hidden SN population is Very Large Telescope (VLT). the CC SN rates give information on the included in the rate calculation, the dis- mass range of their progenitors and on crepancy between SN and star formation the slope of the IMF at the high-mass rates at high seems to disap- The search strategy end. Type Ia SNe, however, are the results pear. It is unclear whether this effect is of thermonuclear explosions of white large enough to also explain the discrep- We conducted the HAWK-I search in dwarfs in binary systems, and they trace ancy observed in the local Universe. K-band, monitoring a sample of local

30 The Messenger 155 – March 2014 starburst galaxies retrieved from the IRAS - - Figure 2. K-band finding Revised Bright Galaxy Sample. We charts for two SNe dis- covered in our search selected galaxies with z < 0.07 and total (left: SN 2012hp; right: 11 infrared luminosity of LTIR > 10 LA. SN 2011ee). The insets We collected a sample of 30 starburst show the transients as they appeared in the dif- galaxies for Paranal observation, consid- $ $ ference images. ering seasonal observability. Given that the SN IR light curves evolve relatively slowly, remaining within one or two mag- nitudes of maximum for two or three ǥ months, an IR SN search did not require ǥ frequent monitoring. We planned on ­average three visits per galaxy per residuals that correspond to the galaxy by the location of the transient inside the semester, which implies a total of about nuclear regions, in particular when seeing host galaxy. Indeed, the magnitude limit 100 observing blocks per semester. was poor (full width at half maximum is brighter in the nuclear regions for a [FWHM] > 1 arcsecond; see Miluzio et al., typical galaxy of our sample, correspond- The programme was allocated three 2013). ing to the inner 1.5–2.0 kpc. While the observing periods and the fraction of detection limit in the galaxy outskirts is useful observing time was 100% of the largely independent of the seeing of the allocated time in the first season and Supernova discoveries and characterisa- images (typically K ~ 19 mag), this is 70% in the second and third semesters. tion not true in the nuclear region, where the In total, we obtained 210 K-band expo- magnitude limit is significantly brighter sures (with a single exposure time of During our monitoring campaign six when the seeing is poor (in the worst 15 minutes). Spectroscopic confirmation ­transients were detected in at least two case even 5–6 mag brighter than in the of the SN candidates was obtained with consecutive epochs separated by around galaxy outskirts; see Miluzio et al., 2013). ISAAC and X-shooter at the VLT (and in one month. We obtained spectroscopic one case at the Gran Telescopio Canarias confirmation for four of them: three as [GTC]). The average image quality was CC SN (SN2010bt; SN2012hp, see SN search simulation excellent: for about 90% of the expo- CBET2446; and SN2011ee, see CBET sures the seeing was less than 1.0 arc- 2773) and one as a type Ia SN (SN To evaluate the significance of the seconds, with an average value across 2010gp). Two candidates were too faint detected events, we developed a simu­ the whole programme of 0.6 arcseconds. for spectroscopy and the transient lation tool that computes the number and Data reduction was performed by inte- ­classification was based on analysis of properties of the expected events based grating EsoRex (the ESO Recipe EXecu- multicolour photometry. We argued that on specific assumptions and features tion tool) with custom programs. Figure 1 they are also likely CC SN. SN e2010bt of our SN search. The basic ingredients shows the high quality of an image of and 2010gp were discovered by optical of the simulation are: the galaxy NGC 6926 (one galaxy of our searches before our detection, but we – the properties of the starburst galaxies sample) after reduction. SN detection rediscovered them independently (Miluzio of our sample; was performed by subtracting images et al., 2013). Figure 2 shows the K-band – the expected SN properties; taken at different epochs with the tech- HAWK-I discovery images of SN 2010hp – the details of the search campaign; nique of point spread function (PSF) and SN 2011ee found in MCG-02-01-52 – the number of SNe expected from the matching. In many cases, the difference and NGC 7674, respectively. The inset stellar population after a star formation images showed significant spurious box is a zoom of the difference image episode; where we detected the SN. – the depth and distribution of dust extinction inside the parent galaxies; – the star formation location in the parent Search detection limit galaxies.

In order to derive the SN rate from the The tool uses a Monte Carlo approach number of detected events, it is crucial that simulates the stochastic nature of to obtain an accurate estimate of the SN explosions. Based on the analysis of detection efficiency of the search. several Monte Carlo experiments with Through artificial star experiments, we the same input parameters, we can test estimated the magnitude detection limit whether the observed events are within for each of the search images and for the expected distribution. On the other ­different locations in the images. The hand, to test the influence of specific detection efficiency is influenced by the assumptions, we varied some of the input Figure 1. K-band HAWK-I image of the NGC 6926 recorded with seeing of 0.4 arcseconds sky conditions at the time of observations parameters in the Monte Carlo experi- (from Miluzio et al., 2013). (namely seeing and transparency) and ments.

The Messenger 155 – March 2014 31 Astronomical Science Miluzio M., Search for Supernovae in Starburst Galaxies with HAWK-I

Observed vs. expected SN discoveries covery and the simulated vs. observed adopted distribution propagates with an extinction distributions are in good agree- error of about 50% on the expected SN Outcomes of the simulations are the ment. This agreement confirms the con- number (Miluzio et al., 2013). expected number of SN discoveries, their sistency of our estimates of the magni- types, magnitudes, extinctions and posi- tude detection limit and the assumption tions inside the host galaxies. In Figure 3 that the extinction is very high in the The future of IR SN searches we show the histogram of the number nucleus and rapidly decreases with the of expected SNe compared to the num- galaxy radius, following the same trend While continuing to search for SNe in ber of observed events (dashed vertical as the SFR. starburst galaxies in the optical and infra- line). Each bin represents the predicted red can certainly help to improve the probability of observing the specific num- We also analysed the distribution of the still poor statistics, one may argue at this ber of actual SNe discovered and the locations of the expected events for an point for a change in strategy. In this shaded areas shows its 1σ Poissonian ideal case, where the magnitude detec- respect, it is worth mentioning the attempt uncertainty range. tion limit in the nuclear regions is as to reveal the hidden CC SNe through deep as in the outskirts, and extinction infrared SN searches that exploit adap- We found that, with the consolidated sim- is negligible. The experiment shows that tive optics at large telescopes, e.g., ulation scenario and input parameters, the fraction of events that remain hidden ­Gemini or the VLT. The early results are we could predict the discovery of 5.3 ± to our search in the galaxy nuclear encouraging, with the discovery of two 2.3 SNe on average, of which 5.1 are CC regions, due to the combined effect of SNe with very high extinction (SN 2004ip

SNe and 0.2 are type Ia SNe (Miluzio et reduced search efficiency and extinction, with AV between 5 and 40 mag, and SN al., 2013). This is in excellent agreement is very high, about 60%. The number 2008cs with AV of 16 mag) and another with the observed number of six events, of CC SNe found in starburst galaxies is two SNe discovered very close to the one of which is a type Ia SN. If we had consistent with what is predicted from galaxy nucleus, although in these cases computed the expected number of SNe the high SFR (and the canonical mass with low extinction. On account of the in our survey based on the cosmic range for the progenitors) when we rec- need to monitor one galaxy at a time and ­average of the SN rate per unit B-band ognise that a major fraction of the events to access heavily subscribed large tele- luminosity or mass, we would predict remain hidden in the inaccessible star- scopes, this approach cannot promise a the discovery of only 0.5 events. The burst regions. This has important conse- large statistical sample, although even a observed number is one order of magni- quences for the use of CC SN as a probe few events would be valuable, mostly for tude higher, which is consistent with the of the cosmic SFR, because the fraction exploring the innermost nuclear regions. fact that the thermal IR emission of star- of SBs is expected to increase with red- On the other hand, a new opportunity burst galaxies is about ten times higher shift (Miluzio et al., 2013). This conclusion that should be exploited is to piggyback than for average galaxies with the same appears to agree with that of previous, on wide-field extragalactic surveys by the B-band luminosity (Miluzio et al., 2013). similar searches (e.g., Mannucci et al., next generation of infrared facilities, in 2003; Mattila et al., 2012). particular the Euclid mission. This would As a consistency check, we verified allow, for the first time, IR SN searches that the distribution of the expected and Finally, to understand how a different to be performed on a large sample of observed apparent magnitudes at dis- choice of the test parameters can influ- ­galaxies, exploring a range of star forma- ence the expected rate, we varied each tion activity and, by monitoring galaxies

  of them and computed the rate, repeat- at different redshifts, probing the cosmic ing the Monte Carlo simulation. One of evolution of SN rates. the main sources of uncertainty is related to the estimate of the magnitude detec-   tion limit, which can vary the predicted References number of SNe by about 15%. The other Botticella, M. T. et al. 2008, A&A, 479, 49 source of uncertainty is the assumption Horiuchi, S. et al. 2011, ApJ, 738, 154   of the CC SN progenitor mass range: Hopkins, A. M. & Beacom, J. F. 2006, ApJ, 651, 142 in particular considering that a different Maiolino, R. et al. 2002, A&A, 389, 84 1D@KHR@SHNMEQ@BSHNM upper limit (10 M instead of 8 M ) Mannucci, F. et al. 2003, A&A, 401, 519 A A Mannucci, F., Della Valle, M. & Panagia, N. 2006,   results in an expected number of SNe MNRAS, 370, 773 about 30% lower than the reference case Mattila, S. et al. 2007b, ApJ, 659, L9 (5.3 SNe). Moreover, while the uncertainty Mattila, S. et al. 2012, ApJ, 756, 111 on the extinction does not significantly Melinder, J. et al. 2012, A&A, 545, A96   Miluzio, M. & Cappellaro, E. 2010, Central Bureau      -TLADQNECDSDBSDC2-D affect the simulation (because where the Electronic Telegrams, 2446, 1 extinction is high our search is limited Miluzio, M. et al. 2011, Central Bureau Electronic Figure 3. Histogram showing the number of by the bright magnitude detection limit), Telegrams, 2773, 1 expected SNe from our Monte Carlo simulations. the most uncertain input of the galaxy Miluzio, M. et al. 2013, A&A, 554, A127 The dashed vertical line indicates the number of Richmond, M. W., Filippenko, A. V. & Galisky, J. observed events and the grey area its 1σ ­Poissonian characterisation is the spatial distribution 1998, PASP, 110, 553 uncertainty (from Miluzio et al., 2013). of the star formation; uncertainty in the Smartt, S. J. 2009, ARA&A, 47, 63

32 The Messenger 155 – March 2014