Supernovae in the nuclear regions of starburst galaxies

Abstract

The feasibility of using near-infrared observations to discover supernovae in the nuclear and circumnuclear regions of nearby starburst galaxies is investigated. We provide updated estimates of the intrinsic core-collapse supernova rates in these regions. We discuss the problem of extinction, and present new estimates of the extinction towards 33 supernova remnants in the starburst galaxy M 82. This is done using H i and H₂ column density measurements. We estimate the molecular to atomic hydrogen mass ratio to be 7. 4±1. 0 in M 82. We have assembled near-infrared photometric data for a total of 13 core-collapse supernovae, some unpublished hitherto. This constitutes the largest database of infrared light curves for such events. We show that the infrared light curves fall into two classes, ‘ordinary’ and ‘slowly declining’. Template JHKL light curves are derived for both classes. For ordinary core-collapse supernovae, the average peak JHKL absolute magnitudes are −18. 4, −18. 6, −18. 6 and −19. 0 respectively. The slowly declining core-collapse supernovae are found to be significantly more luminous than the ordinary events, even at early times, having average peak JHKL absolute magnitudes of −19. 9, −20. 0, −20. 0 and −20. 4 respectively. We investigate the efficiency of a computerized image subtraction method in supernova detection. We then carry out a Monte Carlo simulation of a supernova search using K-band images of NGC 5962. The effects of extinction and observing strategy are discussed. We conclude that a modest observational programme will be able to discover a number of nuclear supernovae.

techniques: image processing, astronomical data bases: miscellaneous, supernovae: general, galaxies: starburst, infrared: galaxies

1 Introduction

Core-collapse supernovae (SNe) are observed to occur in sites of recent star formation. Such regions contain large quantities of dust, especially the nuclear (r ≲ 500 pc) regions of starburst galaxies. Consequently, SN search programmes working at optical wavelengths probably miss a significant number of events in starburst galaxies due to dust obscuration. Thus, the observed supernova rates may be only a lower limit for the true supernova rate. The inferred star formation rates and radio observations of young supernova remnants (SNRs) indicate that highly extinguished supernovae should exist in the nuclear regions of nearby starburst galaxies. In NGC 253 (e. g. Ulvestad & Antonucci 1997), NGC 2146 (Tarchi et al. 2000), M 82 (e. g. Allen & Kronberg 1998) and NGC 4038/39 (Neff & Ulvestad 2000), several compact non-thermal radio sources have been identified as young SNRs, indicating explosion rates of obscured supernovae of around 0. 1 yr⁻¹. Also, high-resolution radio observations have recently revealed a group of luminous radio supernovae in the nuclear regions of ARP 220 (Smith et al. 1998). Yet only one supernova event (SN 1940E in NGC 253) has ever been directly observed in (or in front of) the nuclear regions in any of these five galaxies.

It is desirable to discover and study SNe in the centres of nearby starburst galaxies. In addition to providing a better estimate of the core-collapse rate in the local Universe, such a survey would offer an opportunity to study the behaviour of supernovae in the dusty and dense nuclear environment and to probe the nuclear extinction. Indeed, SNe exploding in the nuclear regions of galaxies may differ significantly from ordinary SNe as a result of the higher-density environment as suggested in the starburst model for active galactic nuclei (AGN) (Terlevich et al. 1992). The best-studied example of a core-collapse SN exploding within a high-density circumstellar environment is SN 1998S (Fassia et al. 2000, 2001; Gerardy et al. 2000; Leonard et al. 2000). Ultimately, such a survey would also be valuable in guiding attempts to determine the rates of high-redshift core-collapse supernovae. Such a high-redshift study is one of the aims of the NGST (New Generation Space Telescope) (Jorgensen et al. 1997; Madau, Della Valle & Panagia 1998; Dahlén & Fransson 1999; Sullivan et al. 2000). However, before starting to investigate the evolution of the supernova (SN) rate with redshift, it is important to have an accurate measurement of the local SN rate and to understand the behaviour of supernovae within the dusty, high-density starburst regions of galaxies.

The feasibility of searching for obscured supernovae in nuclear starburst regions has been discussed in two papers. Van Buren & Norman (1989) considered the use of mid-infrared (mid-IR) narrow-band (10. 52 μm) imaging. They estimated that supernovae would be observable in the mid-IR with ground-based 4–10 m telescopes at distances of 15–40 in one night's integration. Supernova detection in the K band was studied by Grossan et al. (1999). They emphasized the need for high spatial resolution to allow successful image subtraction in the nuclear regions of galaxies.

An optical search for supernovae in a sample of 142 nearby star-forming galaxies was carried out in 1988–1991 by Richmond, Filippenko & Galisky (1998). The observations were carried out with a 500 × 500 charge-coupled device (CCD) camera (0. 58 arcsec pixel⁻¹) on a 1-m telescope at the Lick Observatory. Two search procedures were used. In the first method, the search images were compared by eye. This yielded five supernovae, all of which were outside the extinguished nuclear regions of the host galaxies. The deduced SN rates were similar to those measured in normal galaxies. In the second method, differential photometry was carried out on the nuclei of the galaxies. However, they did not discover any brightening attributable to nuclear supernovae. From this, they deduced upper limits for the unobscured SN rates within the nuclear regions, viz. <9 h² SRU,¹ <12 h² SRU and <7 h² SRU for Types Ia, Ib/c and II, respectively. This work confirmed that, given the likely high obscuration in nuclear starbursts, for a search for supernovae in such regions to be successful it would have to be carried out at infrared wavelengths.

Van Buren et al. (1994) conducted a K-band survey for supernovae in starburst galaxies at NASA's Infrared Telescope Facility, Hawaii, for about 2 yr. Most of the observations were made with PROTOCam, and the Richardson—Lucy image restoration algorithm was used to enhance the resolution of the survey images. They discovered SN 1992bu in NGC 3690 in K-band images. ² The supernova is located ∼6 arcsec from Core B1 of the galaxy. They measured K-band magnitudes of +16. 6, +17. 2 and +18. 1 in three images separated by about a month. No other observation of this supernova was reported, and thus it is not possible to say if it was located within the obscured nuclear regions.

A more recent (1992–1994) near-infrared (NIR) search for supernovae in 177 nearby IRAS galaxies has been carried out in the K′ band by Grossan et al. (1999). The observations were performed with the 2. 3-m Wyoming IR Observatory (WIRO) telescope using the Michigan IR camera (MIRC), with a 2. 2 arcsec pixel⁻¹ resolution and 128 × 128 pixels. The seeing was typically between 0. 7 and 1. 0 arcsec, and so the seeing disc was usually undersampled. The length of time between two observations of a given galaxy varied typically from one to three months. They did not discover any supernovae during the period, limiting the SN rate outside the nuclear regions (>15 arcsec radius) to less than 3. 8 FIRSRU³ in their sample galaxies within 25 Mpc distance. They concluded that this negative result was due to the poor resolution of the camera and that a higher-resolution NIR supernova search covering only the inner ∼450 pc of each galaxy would be more productive.

The latest reported attempt to detect SN explosions in the nuclear regions of starburst galaxies was by Bregman, Temi & Rank (2000). They observed a sample of 10 galaxies within 7 Mpc distance using ISOCAM with a pixel size of 3 arcsec. After continuum subtraction they looked for traces of [Ni ii] 6. 63-μm line emission, which would have indicated recent SN explosions. However, they did not detect any [Ni ii] emission in the sample galaxies. They presented an upper limit for the SN rate in their galaxy sample normalized to M 82, viz. 0. 3 yr⁻¹. Given the far-infrared (far-IR) luminosity of M 82 (Appendix A) this corresponds to 7. 2 FIRSRU.

Although, as argued above, the discovery of such SN events would clearly be important, given the uncertainties in e. g. nuclear extinction or SN magnitudes it is less obvious what might be achieved in practice with existing telescopes. We have therefore carried out a study to test the feasibility of discovering obscured supernovae in the nuclear regions of starburst galaxies. In this paper we report the results of this study. In Section 2, we review indirect methods of assessing the nuclear supernova rate and provide an updated estimate of this parameter. In Section 3, extinction towards nuclear supernovae is discussed. In particular, we present a new estimate of the extinction and its distribution in the starburst galaxy M 82. In Section 4, we present JHKL-band light curves for a total of 13 events, some unpublished hitherto. This constitutes the most complete data base of core-collapse SN IR light curves ever published. We show that the IR light curves fall into two classes, ‘ordinary’ and ‘slowly declining’. Template JHKL light curves are derived for both classes. In Section 5, supernova detection using an image subtraction method is described and tested, and in Section 6 Monte Carlo simulations are used to test the feasibility of a practical supernova search. The results are summarized in Section 7.

2 Indirect estimation of the nuclear supernova rates

A wide variety of indirect methods have been employed to estimate the nuclear SN rates for nearby starburst galaxies. These include (a) observations of young SNRs using radio interferometry, (b) measurement of NIR [Fe ii] luminosities of nuclear regions, and (c) measurement of non-thermal radio luminosities of nuclear regions. A collection of SN rate estimates from the literature is given in Table 1 for M 82, NGC 253 and NGC 4038/39.

Table 1.

Indirect estimates of starburst nuclear supernova rates.

Before examining these rate estimates, we note that the SN rates derived from optical search programmes are often given in terms of the SRU, which has units of number of supernovae per century per 10¹⁰ L_⊙ galaxy blue luminosity (e. g. Cappellaro, Evans & Turatto 1999). However, for nuclear starburst regions the high extinction means that definition of the SN rate in terms of the observed galaxy blue luminosity is rather pointless, since we cannot measure the intrinsic blue luminosity in the region of SN occurrence. Most of the optical—ultraviolet energy emitted by the massive stars is absorbed and reradiated by dust in the far-IR. We therefore express the SN rates in terms of the galaxy far-IR luminosity (i. e. we use FIRSRU).

We can compare directly the estimated SN rate with L_FIR. From Table 1, for M 82, NGC 253 and NGC 4038/39 we adopt SN rates of 0. 1, 0. 05 and 0. 2 yr⁻¹ respectively as representative of the range of estimates. If we then compare these values with the L_FIR values for the galaxies (see Appendix A), we obtain a relation between the SN rate and far-IR luminosity:  

(1)

The origin of this relation is now discussed. The core-collapse SN rate can be derived from the star formation rate (SFR) (M_⊙ yr⁻¹) (e. g. Madau et al. 1998) using  

(2)

where φ(m) is the initial mass function (IMF) with lower and upper mass cut-offs of m_min and m_max, and where m_l and m_u are the lower and upper mass limits for a core-collapse SN progenitor. For SN progenitor masses between 8 and 50 M_⊙ (Tsujimoto et al. 1997) and a Salpeter IMF (m^{−2. 35}) with cut-offs of 0. 1 and 125 M_⊙ (i. e. stars of all possible masses are being formed), we obtain  

(3)

Thus, if we can estimate the SFR in a starburst galaxy, we shall have an additional method for assessing the supernova rate. The SFRs can be estimated for a sample of galaxies via integrated emission-line or continuum luminosities. A popular method for estimating the SFR is via the far-IR luminosity, L_FIR, of the galaxy. This requires us to assume that the L_FIR is powered by a population of young, massive stars rather than by a population of old stars or an AGN. A number of authors have derived SFRs in this way, the estimates for the ratio SFR/L_FIR being in the range (e. g. Thronson & Telesco 1986; Condon 1992; Buat & Xu 1996).

More recently Rowan-Robinson et al. (1997) indicated a relation between SFR and far-IR luminosity using starburst models for optical—UV radiation assuming that a fraction ε of the optical—UV energy emitted in a starburst is absorbed and reradiated by dust in the far-IR. If stars of all possible masses are being formed according to the Salpeter IMF, we have  

(4)

where L_FIR = L(8–1000 μ.)⁴ The choice of the IMF cut-offs is not critical here, as a change in the SFR resulting from a change in the cut-offs is mostly cancelled when converting to the SN rate. This is because the stars exploding as core-collapse SNe are the same stars that produce the luminosity from which the SFR is estimated.

Applying this SFR estimate to equation (2) we obtain a relation between the supernova rate and the far-IR luminosity of a starburst galaxy:  

(5)

For starburst galaxies with high optical depths, ε is about 1. An example is Arp 220, whose LFIR = 1.4 × 10¹² L_⊙ would therefore imply a supernova rate of 1. 5 yr⁻¹. However, for galaxies like NGC 253, M 82 and NGC 4038 the value of ε is smaller (e. g. Silva et al. 1998). Given their L_FIR values of 4.2 × 10¹⁰, 1.9 × 10¹⁰ and 6.8 ×10¹⁰L_⊙ respectively, this implies SN rates exceeding 0. 046, 0. 021 and 0. 075 yr⁻¹, consistent with the typical values given in Table 1.

In summary, equation (1) based on direct observations of SNRs is consistent with equation (5) and agrees well with the r_SN-L_FIR relation obtained similarly by Van Buren & Greenhouse (1994). It is also similar to the unobscured Type (2.5 ± 0.3) × 10⁻¹² × L_FIR/L_⊙ yr⁻¹ SN rate of obtained by Cappellaro et al. (1999). However, the typical ages of starbursts, 10–100 myr (e. g. Genzel et al. 1998), are comparable with the 5 myr lifetimes of the core-collapse SN progenitors. Therefore, the relation between the SN rate and the galaxy's far-IR luminosity depends on the age of the burst (fig. 13 in Genzel et al. 1998), rather than remaining the same from one starburst to another. Starburst models indicate ages of 20–30 Myr for the bursts in M 82 (Efstathiou, Rowan-Robinson & Siebenmorgen 2000) and NGC 253 (Engelbracht et al. 1998). Thus, for a younger or much older starburst, the SN rate would be somewhat smaller. Equation (1) will be used later in this paper to estimate the SN rate for a sample of starburst galaxies.

3 Extinction towards nuclear supernovae

The light from supernovae in starburst galaxies is scattered and absorbed by dust located both locally within the star formation regions in which the SNe occur, and at greater distances but still within their host galaxies. Near-IR recombination lines and colours are widely used for deriving the extinction towards starbursts because they suffer a lower extinction than optical lines. For example, Engelbracht et al. (1998) find A_v = 4–19 towards the nucleus of NGC 253. However, even near-IR lines may not probe deeply enough into these regions. Hydrogen ionization rate studies of Puxley et al. (1997) based on NIR and millimetre-wavelength observations, and mid-IR line studies by Genzel et al. (1998), indicate considerably higher extinctions (A_v = 30–35) towards the NGC 253 nucleus. In Table 2 extinction estimates to galactic nuclei obtained by a range of techniques are listed for several galaxies. In estimating the extinction, two extreme models are often considered, viz. a foreground dust screen and a mixture of stars/gas and dust. These tend respectively to under- and overestimate the extinction. A_v(sereen) ∼ 10 corresponds to A_v(mixed) ≥ 100 (Genzel et al. 1998). It is also likely that the extinction is very patchy, adding further to the uncertainty in the true value. To reduce these uncertainties, we have carried out a new study of the extinction in M 82. This is now described.

Table 2.

Extinction estimates for the nuclei of starburst galaxies.

3.1 Extinction towards the SNRs of M 82

An estimate of the extinction towards the SNRs in M 82 can be made if we know the total hydrogen (atomic+molecular) column densities to these objects. Wills, Pedlar & Muxlow (1998) measured 21-cm atomic hydrogen absorption towards 33 individual SNRs in M 82, from which they derived H i column densities. Weiß et al. (2000) have used large velocity gradient (LVG) calculations to obtain molecular hydrogen (H₂) column densities from high-resolution ¹³CO and C¹⁸O observations for several lines of sight through M 82. By interpolation between these values they have provided estimates of the H₂ column densities through M 82 at the SNR positions (Weiß, private communication). In Table 3 (upper section), columns 3 and 4 respectively, we give the column densities of atomic hydrogen towards the SNRs, N(H i)_SNR, and of molecular hydrogen through the whole galaxy along the line of sight (LOS) to the SNRs, N(H₂)_LOS. Using these values and assuming a conservative error of ±50 per cent for the interpolated N(H₂) values, we obtain the ratio for each SNR. These are presented in Table 3, column 5. We expect this ratio to have its smallest values for the SNRs that are located behind most of the molecular gas along the line of sight, i. e. the minimum values give us an indication of the mass ratio of H₂ to H i along a given line of sight.

Table 3.

Extinction estimates for the SNRs of M 82. The right ascensions and declinations are offsets from 09^h51^m and +69^d54^m00^s (1950.0).

Uncertainties in the ratios are due to errors in the N(H i) values, the CO data entering the LVG calculations and the interpolations to the SNR positions. In order to make better use of the column density data, we follow the philosophy of the V/V_max test (Schmidt 1968). In this we assume that has a single value for any line of sight through M 82. In addition, we assume that the SNRs are randomly distributed throughout M 82, and so we expect to have  

(6)

Applying the test using the 19 sets of H i (excluding the upper and lower limits) and H₂ column densities (Table 3), we find an H₂ to H i mass ratio of 7.4 ± 1.0. However, it should be noted that the H₂ column densities are model-dependent, introducing further uncertainty in the mass ratio. Nevertheless, our result is consistent with the lowest values listed in Table 3, but has a higher precision. It is also in good agreement with the mass ratios of 7–10 suggested by Crutcher, Rogstad & Chu (1978) and Yun, Ho & Lo (1993) for the centre of M 82.

The extinction towards the M 82 SNRs was estimated adopting (Bohlin, Savage & Drake 1978). This assumes that the dust properties and dust/gas ratio are the same for M 82 as for the Milky Way Galaxy. For each SNR, an estimate of A_V was obtained by adopting 7. 4 for the 2N/(H₂)/N(Hi) ratio and using the N(H i) values. The extinctions are listed in Table 3, column 6. The uncertainties arise from errors in the N(H i) measurements and in the adopted 2N/(H₂)/N(Hi) ratio. We note that the SNRs with the highest extinctions are located within the two CO lobes on the edges of the galaxy. It has been suggested that these lobes are associated with the ends of an edge-on ring structure or bar structure (Neininger et al. 1998). Considering only those 19 SNRs for which we have N(H i) values (as against lower or upper limits), we obtain an unweighted mean extinction of A_v = (σ ∼ 9). However, M 82 is an almost edge-on galaxy and so, in general, the column densities will be considerably greater than for a face-on starburst galaxy. Incorporation of our M 82 extinction analysis into the feasibility study will be described in Section 6.1.

4 The infrared light curves of core-collapse supernovae

In order to investigate realistically the feasibility of discovering nuclear SNe, we need to ascertain the IR luminosity and evolution of core-collapse SNe. To this end we have assembled near-IR photometric data for a total of 14 events, some of which have never been published. This constitutes the most complete data base of core-collapse SN IR light curves to date. However, one event was SN 1987A, and although coverage of its IR light curves is unparalleled, they were highly atypical in luminosity and shape. We have therefore excluded this event. The photometric data for the remaining 13 SNe are listed in Appendix B. In Table 4 we list other information about these SNe. We have used these events to derive absolute magnitude ‘template’ light curves to represent the typical core-collapse SN evolution in the IR.

Table 4.

The supernova sample.^a

Distances to the host galaxies are listed in column 5. For SN 1979C (Ferrarese et al. 1996) and SN 1993J (Freedman et al. 1994) these were obtained from HST Cepheid observations. For SN 1980K, Eastman, Schmidt & Kirshner (1996), Buta (1982) and Tully (1988) give similar distances using, respectively, the expanding photosphere method, photometric distance indicators, and recession velocities corrected for the expansion of the Local Supercluster and the infall towards the Virgo cluster. We adopted the average of these three values. For SN 1990E, we adopted the average of the similar distances given by Eastman et al. (1996) and Tully (1988). For the remaining nine SNe, we used the distances given by Tully (1988). ⁵ Estimates of the errors in the distances are given in parentheses.

For seven events (SNe 1979C, 1980K, 1983N, 1990E, 1993J, 1994I and 1998S) photometry is available for at least one of the JHKL bands less than ∼10 d after maximum light in the BV band. ⁶ Extinction estimates for these SNe are also available from the literature, and these are listed in column 6. These values were converted to A_J, A_H, A_K and A_L assuming the Galactic reddening law given by Rieke & Lebofsky (1985). Absolute intrinsic magnitudes near epoch 0 d were then determined by subtracting the distance moduli from the apparent magnitudes and correcting for extinction. These magnitudes are listed in columns 7–10. The other six events (SNe 1980E, 1980R, 1983I, 1984L, 1985L and 1985P) were not observed until an epoch of later than 10 d. Moreover, no extinction estimates were available for these events. Therefore, the extinctions were estimated by adopting a typical host-galaxy extinction of A_v = 0.7 and adding to this the appropriate Galactic extinction value given in Schlegel, Finkbeiner & Davis (1998). The uncertainty in A_V is therefore quite large – we estimate ±0. 5. The peak magnitudes for these SNe given in Table 4 (columns 11–14) are only estimates, obtained by extrapolating back to epoch +5 d for ‘ordinary’ events and to ∼ +10 d for ‘slowly declining’ events using the light-curve templates described below.

Among the 13 events used to determine the absolute magnitude template light curves, we identify two classes of behaviour. In one class we have four supernovae (including Types IIP, IIL, Ib) which decline linearly for at least ∼200 d after epoch 0 d. We shall refer to these as ‘ordinary’ events. This class includes SNe 1980K, 1982E, 1983N and 1985P. In the other class we have two ‘slowly declining’ supernovae, comprising the Type IIL SN 1979C and the Type IIn SN 1998S. From about 0 d to 100 d the JHK light curves of this class decline at a rate similar to that seen in the ordinary light curves. However, beyond ∼100 d a much slower decline in J and H occurs, together with virtually no decline in K, persisting for over 450 d. Moreover, in the L band there is negligible decline from soon after the explosion to beyond 450 d. Recent observations show that the slowly declining behaviour has persisted in SN 1998S for over 700 d. The remaining seven events (SNe 1982R, 1983I, 1984L, 1985L, 1990E, 1993J and 1994I) all show a similar decline rate during the first 100 d to that seen in the ordinary and slowly declining classes. They include Types IIP, IIL, IIb and Ib/c. However, as they were not observed in the IR much beyond 100 d, we are unable to distinguish which of the two classes, if either, they should be assigned to. Therefore, to be conservative in the simulations, we assign them to the ordinary class.

We constructed absolute magnitude template light curves for the two classes. This was done as follows. Inspection of the ordinary individual JHK light curves suggested that two-component linear templates might provide a reasonable description of the data. We therefore fitted functions of this form to the entire ordinary absolute magnitude data set in each band. Epoch +5 d was selected by visual inspection as a plausible intersection point for the two components. The template fit for the data after +5 d was then obtained in two stages. In the first step, the best-fitting slope was determined by iteratively adjusting the vertical position of the individual SN light curves. The reason we followed this procedure was to minimize bias in the slope, which would otherwise be introduced by differences in temporal coverage for different SN events having different peak magnitudes. As we were only interested in the slopes at this stage, only the photometric errors were included in the fits. Reduced χ² values of 20, 10 and 6 were obtained in JHK respectively. The larger-than-unity values of the reduced χ² values obtained indicates some intrinsic variations in light-curve shapes between SN events. In the second step, the absolute peak magnitudes for the templates were obtained. For each band, the post +5 d template slope was fixed at the value obtained in the procedure just described. The templates were then compared with the original, unshifted data points of individual supernovae. The vertical position of the post +5 d light curve of each individual supernova was adjusted to obtain the best fit to the points, i. e. the only free parameter was the absolute magnitude position of the light curve. The absolute peak magnitude of the template was then determined by taking a weighted average of the individual supernova peak magnitudes (Table 4, columns 11–14) The weights for the events were obtained by combining the error in the apparent peak magnitude obtained in the fit with the errors in the distance modulus (±0.2–0.3 mag) and in the extinction corresponding to uncertainties in A_V of ±0.1–0.5 (see Table 4). Turning now to the pre +5 d template component, the +5 d magnitude of this component was set to match that of the post +5 d component. The slope of the pre +5 d component was then varied to provide the best fit. The fits were carried out after shifting the data points vertically so that the peak magnitudes of the individual events and the template were the same. Only the photometric errors were included in the fit. Reduced χ² values of 50, 86 and 63 in JHK respectively were obtained, indicating a considerable variation in the pre-maximum light-curve slopes. For the L data, a single linear fit provided an adequate description. As with the JHK bands, the best slope was found by iteratively adjusting the vertical position of the individual SN light curves. This yielded a reduced χ² of 1. 0. Again, the vertical position of the template was determined by a weighted average of the absolute magnitudes of the individual events obtained in the same way as before.

The JHKL ordinary template parameters are given in Table 5. In a given band, M(5) is the absolute magnitude of the template peak at +5 d; δ₁ and δ₂ are the light-curve slopes before and after epoch +5 d. The decline rates ranged from 0. 027 mag d⁻¹ in J to 0. 013 mag d⁻¹ in L. In the bottom row we give the weighted dispersion, σ_y, of the individual supernova peak magnitudes about the peak of the template. The ordinary templates and individual data points are shown in Fig. 1, where the horizontal position (epoch) is with respect to the epoch of BVmaximam = 0 d (see footnote 6). The error bars on the data points are the combined errors in photometry, extinction and distance.

Table 5.

Template light curve of ordinary SNe.

Figure 1.

JHKL-band light curves and templates for ‘ordinary’ core-collapse SNe. For clarity, the J-, H- and L-band data and templates have been vertically displaced by +10, +5 and −4 mag respectively. The horizontal positions of the individual light curves were set such that 0 d corresponds to BV maximum (see text) in each case. The error bars give the combined uncertainties in photometry, extinction and distance.

For the two slowly declining SNe, we adopted three-component linear templates in JHK, with a single-component linear template in L. Guided by visual inspection, we divided the data into three eras corresponding to pre +10 d, +10 to +100 d and post +100 d. Linear fits were then carried out for each component. Only the errors in photometry were included. The best-fitting slope for the epoch between 10 and 100 d was obtained again by iteratively adjusting the vertical positions of the individual events. This slope was then used for estimating the absolute peak magnitudes of both events, and thus the vertical position of the template as before. The data points of the individual events were then shifted vertically so that their peak magnitudes matched the one of the template. This allowed us to perform linear fits for the pre +10 d and post +100 d components. The magnitudes of the points of intersection (i. e. the template peak and inflection point) were derived from the fits. Reduced χ² values were 41, 29 and 2 for +10 to +100 d and 15, 11 and 14 for post +100 d in JHK respectively. The value for the single-component L-band fit was 2. For the pre +10 d, the reduced χ² values were 4, 10 and 5 in JHK respectively. The template parameters are shown in Table 6. M(1) and t₁ give, respectively, the absolute magnitude and epoch of the template peak, while M(2) and t₂ give the absolute magnitude and epoch of the inflection; δ₁, δ₂ and δ₃ give the respective slopes for the three eras. The bottom row gives the dispersion, σ_y, of the post +10 d points about the template. The templates and individual data points (with the photometry errors) are shown in Fig. 2, where, as before, the horizontal position (epoch) is with respect to the epoch of BVmaximum = 0 d (see footnote 6). As already indicated, we find that up to about +100 d, the decline rates are the same for both classes, to within the errors. However, the defining characteristic of the slow decliners is the slope after +100 d. For this era, we obtain slopes of 0. 0086, 0. 0030, ∼0 and 0. 0018 mag d⁻¹ in the J, H, K and L band respectively. However, our study has revealed an important additional characteristic of the slow decliners, viz. that they are significantly more luminous than the ordinary class, even at early times. Around maximum light SNe in this class are ∼1. 5 mag brighter in JHKL. However, it should be remembered that we have identified only two slow decliners in our sample. Presumably, the slowly declining late-time light curve of this class, and perhaps also the greater early-time luminosity, is being powered by conversion of the SN kinetic energy into IR radiation through shock interaction with the circumstellar medium and heating of dust local to the supernova (see e. g. Fassia et al. 2000, 2001).

Table 6.

Template light curve of slowly declining SNe.

Figure 2.

JHKL-band light curves and templates for ‘slowly declining’ core-collapse SNe. For clarity, the J-, H- and L-band data and templates have been vertically displaced by +8, +4 and −4 mag respectively. The horizontal positions of the individual light curves were set such that 0 d corresponds to BV maximum (see text) in each case. The error bars give the combined uncertainties in photometry, extinction and distance.

Because of the reduced extinction, and the slower light-curve decline rates compared with the J and H bands, the K and L bands are superior for the purpose of a nuclear SN search. Below we examine the feasibility of detecting such SNe using repeated K-band imaging of a sample of nearby starburst galaxies.

5 A technique for supernova detection

Having estimated the intrinsic rate of core-collapse supernova explosions (Section 2), determined the typical range of extinctions towards SNe in starburst galaxies (Section 3), and established the magnitude and shape of K-band SN light curves (Section 4), the next stage is to examine the efficiency with which we might detect SNe in galaxy nuclear regions.

5.1 K-band imaging of NGC 5962

We used an image pair of the IR-luminous galaxy NGC 5962 to study the efficiency of SN detection in the galaxy nuclear regions. NGC 5952 is a star-forming galaxy at a distance of 34 Mpc (Tully 1988) with a far-IR luminosity similar to the prototypical starburst galaxy M 82

State-of-the-art K-band images of NGC 5962 were obtained by A. Fassia, M. Hernandez and T. Geballe using UFTI (0. 091 arcsec pixel⁻¹) at UKIRT. Repeat images were obtained on 1999 April 4 and 5. The seeing was about 0. 6 arcsec on the first night and about 1. 0 arcsec on the second night. Each image comprised a mosaic pattern of five individual frames with 10 or 11 arcsec offsets. The total integration time for a complete image was 400–600 s. For each of the observations a sky flat was observed with an equal exposure time and dither pattern. In addition a 5 ×60 s exposure sky flat was obtained at the beginning of both nights. During the data reduction it was found that the best result for the April 4 image was achieved by using the contemporary sky flat for flat fielding. For the April 5 image, a better result was achieved with the sky flat taken at the start of the night. The images were calibrated using the UKIRT faint standards FS21 and FS25. The images are shown in Fig. 3.

Figure 3.

K-band images of NGC 5962 obtained with the UFTI camera at UKIRT (north is up, and east to the left). The plate scale is 0. 091 arcsec pixel⁻¹. The April 4 image (seeing FWHM = 1.0 arcsec is on the left and the April 5 image (seeing FWHM = 1.0 aresec) is on the right. The circle in the left-hand image is of radius 500 pc (∼3 arcsec). The contour interval is 40 counts. The lowest and highest contours are 50 and 410 counts, which correspond to +15. 9 and +13. 6 mag arcsec⁻² respectively in the left-hand image, and +15. 7 and +13. 5 mag arcsec⁻² respectively in the right-hand image.

5.2 Supernova detection

Supernovae occurring in the nuclear regions of galaxies can be sought by comparing two images taken at different times. A rough comparison can be made by examining the images by eye and varying the contrast levels. A more efficient approach is by ‘blinking’ the two frames. It is likely that the most sensitive search procedure is to carry out computerized subtraction of one image from the other to reveal very small differences. This is the method used in the feasibility study described here. However, for this to be practicable it is necessary first to correct for the effects of differing atmospheric conditions (seeing and transparency), exposure times, focus positions and telescope guiding accuracy between the two images. Ideally, the images should all be taken with the same instrument.

5.2.1 Image matching

Before attempting to subtract one image from another, they must first be aligned accurately. The shifts in x and y are obtainable from the positions of the galactic nucleus and any bright point sources within the images. Owing to the small pixel size of UFTI, the alignment did not need to be performed to a precision of better than half a pixel.

A more difficult problem is image mismatch due to differences in seeing and photometric conditions between the two observations. Initially, we attempted to deal with this by matching the point spread functions (PSFs) and the intensity levels of the two images using the standard iraf⁷ image matching package immatch. However, this was found to give unsatisfactory results in regions around bright stars and galactic nuclei. We therefore turned to the more sophisticated Optimal Image Subtraction method of Alard & Lupton (1998), which they developed for analysis of microlensing survey images. It derives an optimal kernel solution from least-squares analysis of the data. The method has the advantage that it does not require any bright isolated stars to determine the kernel but can be used on any portion of an image with high enough signal-to-noise ratio. In our feasibility study we used a new version of the method (Alard 2000), which was developed to process non-crowded field images. In this version, instead of using all the image pixels, only the regions around selected bright, but not necessarily stellar, objects are used to find the kernel, since in a non-crowded image most of the pixels do not include any useful information for deriving the kernel solution. Among the bright objects we must include the innermost region of the galactic nucleus if we are to obtain a satisfactory subtraction result for the nuclear regions. Unfortunately, this makes the detection of supernovae falling on those innermost pixels impossible. In the images of NGC 5962 the size of the nuclear region involved in the kernel solution was 16 × 16 pixels, making detection of SNe falling within the innermost ∼0. 7 arcsec impossible. However, we estimate that only ∼20 per cent of the supernovae would occur within this region (see Section 6). Both the nucleus and also regions around four brightest point sources (m_K > 16) outside the galaxy's circumnuclear regions were used for the fit. In order to get the best results, all the regions used for the fit must be within the common portion of the mosaic image. Here, all the regions used for the fitting procedure were within 30 arcsec of the nucleus and are thus easily fitted within a single field of view of a typical infrared imager.

5.2.2 Testing the Optimal Image Subtraction method for SN detection

We used the iraf package mkobjects to add artificial SNe to the galaxy images. To set the intensity profile of the artificial SNe, an elliptical Moffat profile with a similar position angle, ellipticity and beta parameter to the real stars in the image was used. This was set to correspond to the better-seeing image (∼0. 6 arcsec) and the artificial supernovae were added to this image. However, since the better-seeing image is convolved to match the poorer-seeing image before the image subtraction, the size of the supernova seeing disc in the residual image is ∼1 arcsec. An example of supernova detection is presented in Fig. 4. In Fig. 4 (left) an artificial supernova of m_K = + 17.0 has been placed 1. 1 arcsec (180 pc) south of the centre of NGC 5962 in the better-seeing image, indicated by the cross. In the residual image in Fig. 4 (right) the supernova is clearly visible.

Figure 4.

K-band image of NGC 5962 (seeing FWHM = 0.6 arcsec) with a simulated SN (cross) of m_K > = 17 at (441, 472) (left) and after image subtraction (right). The plate scale is 0. 091 arcsec pixel⁻¹. The contour interval in the subtracted image is one count, and the zero contour is at two counts, corresponding to +19. 3 mag arcsec⁻². The residual image (right) has been binned by ×5 in the x and y directions.

The occurrence of 18 000 supernovae was simulated, distributed randomly within the innermost 500-pc radius of NGC 5962. For each simulation, subtraction was performed using the procedures described above. The residual images were then sought for point sources using the irafdaophot package. The detection threshold was set at five times the background noise. The supernova detection efficiency at different distances from the centre of NGC 5962 is presented in Fig. 5 for a range of apparent magnitudes. It shows the probability of detecting SNe of different apparent magnitudes and positions. Only in the 0–100 pc region could no supernovae be detected by this method.

Figure 5.

The fraction of simulated SNe recovered within different annuli in the K image of NGC 5962 as a function of the SN apparent magnitude. The dotted, solid, short dashed, long dashed and dashed-dotted lines correspond to the 0–100, 100–200, 200–300, 300–400 and 400–500, annuli respectively.

6 Feasibility of searching for supernovae in the nuclear starburst regions of galaxies

The SN detection efficiency derived above is used here to determine the number of nuclear supernovae that would be discovered in a practical survey.

6.1 Monte Carlo simulations of the appearance of supernovae

We examined the detectability of supernovae in the nuclear regions of starburst galaxies as follows. We simulated the range of absolute peak magnitudes by applying Gaussian distributions to the two types of template light curve. Average peak magnitudes of K = −18.6 and K = − 20.0 and dispersions of σ_K = 0.8 and σ_K = 0.5 were adopted for the ordinary and slowly declining SNe respectively. Since 1987, the fraction of core-collapse supernovae exploding as Type IIn events listed in the Asiago Supernova Catalogue (Barbon et al. 1999) is 14 per cent. However, the intrinsic rate of Type IIn SNe may be smaller since their higher intrinsic brightness compared to the ‘ordinary’ SNe means that they are more likely to be discovered. According to Cappellaro et al. (1997) they constitute only per cent of all Type II SNe. However, as the proportion of slow decliners in the nuclear starburst regions is uncertain, we carried out the simulation for slow decliner fractions ranging between 0 and 60 per cent. The epoch of observation was allowed to vary randomly between −20 and +340 d relative to the K maximum light. The absolute magnitude for each of the events was then assigned according to the template light curves (Section 4).

The absolute magnitudes were then scaled by the appropriate extinctions and distance moduli. To simulate the effects of extinction, we carried out the study for three fixed extinction values, viz. 3 and 4 27 and 36). We also carried out the simulation for a specific distribution of extinction values towards the nuclear SNe, by adopting the extinction distribution towards the SNRs in M 82. This was done by selecting an extinction for each simulated event randomly from the extinctions listed in Table 3, column 6. In the case of the extinction lower limits the SN was assumed to be totally obscured in the K band regardless of its magnitude, whereas the extinction upper limits listed in Table 3 (lower section) were considered as actual extinction values.

We simulated the occurrence of supernovae in galaxies at distances of 20, 30 and 40 Mpc, and also carried out the simulation for a specific distribution of distances according to the galaxy sample described in Appendix A. This was done by selecting the distances for the simulated events from a random distribution in which each of the distances had a weight according to the corresponding galaxy's far-IR luminosity, i. e. its SN rate. As the distance increases, not only does the supernova become fainter but also, for conservatively poor seeing of 1 arcsec, the ability to resolve any SN spatially against the nuclear region rapidly diminishes. One arcsec at 50 Mpc corresponds to 240 pc, which is already half of our search radius. Therefore, we considered the detection of SNe within 45 Mpc only. In considering different host-galaxy distances, we only varied the SN magnitudes. The actual NGC 5962 images were unchanged. For galaxies at distances less than 34 Mpc (the distance to NGC 5962), this is clearly a conservative approach because of the larger linear size of the galaxy nuclear profile compared to the size of the seeing disc. However, for galaxies more distant than NGC 5962 the reverse is true. The linear region over which the SN can be spatially resolved against the nucleus diminishes. The poorer seeing image of the NGC 5962 image pair was 1. 0 arcsec. Thus, to achieve comparable resolution for a galaxy at 45 Mpc, we would need the seeing of the poorer image to be 0. 75 arcsec. We therefore assumed that, in the survey, the best seeing nights would be devoted to the most distant galaxies of the sample.

The intrinsic rate of core-collapse SNe in the nuclear regions of the sample galaxies was scaled according to the galaxy far-IR luminosities, L_FIR (see Appendix A). For the relation between the core-collapse SN rate and L_FIR we adopted the average of that found in NGC 253, M 82 and NGC 4038/39, viz. The supernova location was based on the observation by several authors that the K-band continuum in a starburst system (e. g. NGC 253; Engelbracht et al. 1998) is dominated by red supergiants (the progenitors of core-collapse SNe). We assume that this is roughly appropriate for NGC 5962, which has an L_FIR higher than NGC 253, and is classified as a ‘nuclear H ii region galaxy’ (Misselt, Clayton & Gordon 1999), indicating a moderate level of nuclear star formation. We therefore used the nuclear K-band light profile as being roughly representative of the SN spatial distribution. Random distributions of (x, y) coordinates were generated for the artificial supernovae, weighted according to the observed K-band light distribution of NGC 5962 within 500 pc of the centre.

6.2 Simulation of a supernova search

We used the simulated occurrence of the supernovae together with the supernova detection efficiency estimated above to predict the number of supernovae discoverable during an observing programme using a ∼3-m telescope (e. g. the NASA Infrared Telescope Facility or the Nordic Optical Telescope). We considered a sample of 45 galaxies (see Table A1) observable in the Northern hemisphere. These 45 galaxies were selected according to their far-IR luminosities, far-IR colours and distances (see Appendix A).

Table A1.

The galaxy sample. The flux densities are taken from Soifcr et al. (1989) and Rowan-Robinson & Crawford (1989), and the distances mostly from Tully (1988). For MGC-05-17-009, MGC-05-18-003, ESO 320-G30, ESO402-G26, NGC 3690 and UGC 3630 the distances were obtained from the recession velocities with respect to the Local Group (H₀ = 70kms⁻¹ Mpc⁻¹). For NGC 253 and M 82 the distances were adopted from de Vaucoulcurs (1978) and Tammann & Sandage (1968) respectively, and for NGC 4038/39 the same distance was assumed as in Ncff & Ulvcstad (2000). The far-IR luminosities 1(8-1000 |xm) were calculated from the given flux densities and distances according to Sanders & Mirabel (1996). The far-IR luminosities are converted to the intrinsic SN rates according to (a) equation (5) and (b) equation (1) in Section 2. The values given in column (a) were obtained assuming e= 1 and so are probably lower limits.

In Fig. 6 we show, for a fixed integration time per galaxy, the fraction of simulated SN detected in a single observation as a function of time interval, t_intv, since the previous observation. Each curve is based on the simulation of 100 000 SNe within the innermost 500-pc radius of NGC 5962 (see Fig. 3, left) as described in Section 6.1. The effects of intrinsic dispersion in the supernova magnitudes and a 5 per cent intrinsic fraction of slow decliners have been included. Results for a range of extinctions and distances are illustrated. For example, at 30 Mpc, for a time interval of 200 d between successive observations, we find that if A_V is increased from 18 to 36 the fraction of SNe recovered is reduced by a factor of ∼4. Likewise, for an extinction of increasing the simulated distance to the host galaxy from 20 Mpc to 40 Mpc cuts the fraction detected by a factor of ∼2. 5. These results are now used along with the predicted intrinsic SN rate (cf. Section 2) to examine how the number of discoverable SNe in a galaxy sample is affected by choice of t_intv.

Figure 6.

The fraction of simulated SNe recovered within the innermost 500 pc in the K image of NGC 5962 for different time intervals between observations. The top left, top right and bottom panels correspond to simulated galaxy distances of 30 and 40 Mpc respectively. In each panel, the dashed (a), solid (b) and dotted (c) lines correspond to 27 and 18 respectively.

In Fig. 7 we show the number of supernovae discoverable from the sample of 45 galaxies for a given observation of the entire galaxy sample (assuming that the galaxy comparison images have been already acquired) as a function of the time interval, t_intv, since the previous observation. Again, each curve is based on the simulation of 100 000 SNe. These figures are used below to estimate the best observing strategy. The number of SNe discoverable is plotted in Fig. 7 (left) for three different extinction values assuming a 5 per cent fraction of slow decliners, and in Fig. 7 (right) for the M 82-like extinction distribution (see Table 6, column 6). In Fig. 7 (right) we also illustrate the effect of the expected higher rate of the slowly declining events in the nuclear starburst environment. Such an enhanced rate could be due to the higher densities caused by greater stellar mass loss, stellar winds and other, nearby, SNRs in the nuclear regions (e. g. Cid Fernandes 1997). A recent study of the SNR flux variation in M 82 by Kronberg et al. (2000) gives indirect evidence of enhanced circumstellar densities. They found that only ∼25 per cent of the sources varied measurably during a 11. 8-yr period, with the remaining ∼75 per cent having exceptionally stable integrated flux densities. We conclude that, for months, if about 40 per cent of the nuclear SNe were slowly declining, we would expect a ∼35 per cent higher discovery rate than would be achieved for a ∼5 per cent SNe IIn fraction.

Figure 7.

The total number of SNe discoverable in a single observation of the whole galaxy sample (Table A1) as a function of the length of time since the previous observation, t_intv. In the left-hand plot, the dashed (a), solid (b) and dotted (c) lines correspond to 27 and 18 respectively, with a 5 per cent fraction of slow decliners. In the right-hand plot, we have adopted an M 82-like extinction distribution, and the lines a, b, c, d, e and f correspond to different intrinsic fractions of slowly declining events, viz. 0, 0. 05, 0. 10, 0. 20, 0. 40 and 0. 60 respectively (see Section 6.2).

With limiting magnitudes similar to those of the NGC 5962 images used in this feasibility study, the 45 sample galaxies (and sky flats) could be observed in ∼20 h with a telescope (K band). In a single observation of the sample, the expected number of supernova discoveries is 1. 8, 1. 0 and 0. 5, for extinctions of 27 and 36 respectively. This result corresponds to a time interval exceeding ∼6 months since the previous observation of the sample (see Fig. 7, left). If we assume that the distribution of extinctions towards the SNe is similar to that of M 82, we obtain lower and upper limits of 1. 0 and 1. 3 respectively for the expected number of SN detections, given slow decliner percentage fractions of 5 and 40 per cent respectively (see Fig. 7, right).

The adoption of the extinction distribution based on M 82, which is an edge-on galaxy, our conservative approach to simulating supernovae in galaxies closer than 34 Mpc (see Section 6), and the likely higher fraction of the slowly declining events in the nuclear starbursts, taken together suggest that our supernova discovery rate estimates are probably lower limits.

However, even if no supernovae were discovered in an observing programme, useful upper limits could be derived for the nuclear supernova rate, assuming particular extinction values. For example, in a negative search consisting of three, four, or five observations of the sample, an upper limit for the supernova rate would be 3. 1, 2. 1, or 1. 6 FIRSRU at 90 per cent confidence, assuming an M 82-like extinction (Section 3), and a slow decliner percentage fraction of 5 per cent. Alternatively lower limits could be derived for the average extinction towards the nuclear supernovae, assuming our estimated intrinsic nuclear supernova rate of 2. 7 FIRSRU. For negative surveys comprising three, four and six observations of the sample, the extinction lower limits would be 3 and 4 respectively, with at least 90 per cent confidence

6.3 Observing strategy

Given a fixed amount of telescope time, what distribution of the time will maximize the number of detected supernovae? Clearly, as we increase the integration time per galaxy we shall (a) increase the volume of space within which SNe can be detected, and (b) increase the amount of time after explosion during which we can detect a given SN. However, is it better to apply the same integration time to all galaxies, or to increase the integration time as we go to more distant galaxies? Keeping the integration time the same means that we can detect the much rarer nearby SNe at much later points on their light curves and so reduce the chance of missing these particularly interesting events. The disadvantage is that this procedure will sometimes ‘waste’ time detecting a nearby SN while it is still at an early point on its light curve and therefore very bright. On the other hand, if we adjust the integration time to give a detection threshold which is at the same epoch on the light curve regardless of the galaxy distance, then we can save time on the nearer events and use the time saved to increase the volume of space explored. The disadvantage is that we are more likely to miss the fewer nearby events. It can be shown that the latter strategy will yield a somewhat higher supernova detection rate. However, this ignores the fact that as we go to greater distances, the supernovae become increasingly difficult to resolve against the host-galaxy nucleus. When we take that into account, the benefit of varying the integration time with distance diminishes. This fact, together with the desire to avoid missing nearby SNe, suggests that working with a fixed integration time per galaxy is a reasonable strategy.

We turn now to the question of the number of times we should observe a given galaxy. For a given amount of telescope time and a fixed, uniform integration time (i. e. magnitude threshold) per galaxy, if we double the number of observations it simply halves the number of galaxies we can observe. In another approach, which would allow a doubling of the number of times a given galaxy was observed, we could maintain the number of galaxies observed but reduce the integration time on each by a factor of 2 (ignoring overheads). This would effectively reduce the number of distant SNe that would be detected. The volume of space accessible above the threshold would be reduced by while doubling the number of observations. This would yield a modest net gain in the detection rate of So, given the overheads, there would be nothing gained in going for a larger number of shorter observations. Thus, it appears that there would be little to be gained in dividing up the telescope time into more than a few observations per galaxy.

Another issue concerns the interval between successive observations of a given galaxy, t_intv. Inspection of Fig. 7 reveals that, at first, the SN detection rate increases quite rapidly with t_intv. This is due to the fact that, as t_intv is increased, more supernovae will occur in the interval and yet we are still able to detect many of those SNe that exploded soon after the previous observation. If we work with a fixed detection threshold, then the nearer, but rarer, SNe can be detected at much later points on their light curves. As a rule of thumb, therefore, we should set t_intv to equal the light-curve duration over which we can detect the nearest events. Less time than this will reduce the number of the nearest events we can detect. More time than this does not increase the number of detectable ‘ordinary’ events, but simply increases the duration of the programme. It is true that as we continue to increase t_intv, more slow decliners will be found (see Fig. 7, right). However, the conservative assumption is that these form a small minority of the sample. For an ‘ordinary’ template K-band light curve with M_K peak at −18. 6 and assuming our UFTI images suggest that the nearest likely supernova could be detected out to ∼140 d post-explosion in an integration time of on a telescope. Thus we are looking at t_intv values of at least a few months. Taking into account also the slow decliners, we conclude that telescope time is used most efficiently if the time intervals between the observations of a given galaxy are at least∼6 months.

7 Conclusion

The feasibility of detecting supernovae in the nuclear regions of nearby starburst galaxies has been investigated. This has entailed the examination of expected SN rates, extinction and SN magnitudes in such regions. An intrinsic SN rate of was estimated. Extinction values ranging between and 4 were judged to be typical. A particularly detailed examination of the M 82 extinction distribution was carried out. A mean A_V of 30 was obtained from hydrogen column density measurements. Template core-collapse SN light curves in JHKL were assembled. Two types were identified – ordinary and slowly declining. The slow decliners were found to be significantly more luminous than the ordinary core-collapse SNe, even at early times. Using the Optimal Image Subtraction method, it was found that supernovae outside the innermost ∼1 arcsec of a galaxy can be easily detected in K-band images. Imaging of a sample of 45 galaxies on a six-month cycle would probably reveal at least one supernova per cycle. However, the actual number of detections could be significantly higher if slow decliners are more common in nuclear regions. Near-infrared observations of SNe in nuclear starbursts will provide us with a valuable new probe of the conditions and star formation rates in the obscured starburst regions of galaxies.

Acknowledgments

We are grateful to C. Alard for providing us with the Optimal Image Subtraction program and his generous help in its usage, which made this study possible. We are also grateful to A. Fassia, M. Hernandez and T. Geballe for carrying out the NIR observations of NGC 5962 at the United Kingdom Infrared Telescope (UKIRT), which is operated by the Joint Astronomy Centre on behalf of the UK Particle Physics and Astronomy Research Council. We thank A. Efstathiou, A. Fassia, T. Hawarden, L. Lucy, N. Neininger, M. Rowan-Robinson, S. Ryder, K. Wills, K. Mattila, D. Farrah and O. Prouton for helpful discussions. We also thank N. Neininger and A. Weiß for the N(H₂) values for M 82 prior to publication. We are grateful to the anonymous referee for a number of helpful suggestions. We acknowledge the use of C. Markwardt’s MPFIT IDL programs and programs from Numerical Recipes. Financial support was provided to SM by Jenny ja Antti Wihurin rahasto, Vilho, Yrjö ja Kalle Väisälän rahasto, Osk. Huttusen Säätiö and the British Council. This research has made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

1

One SRU is the number of supernovae per century per 10¹⁰ L_⊙ galaxy blue luminosity, and

2

for further information and/or details see the website: edu/ipac/info/sn1992/sn1992_NGC3690. html.

3

One FIRSRU is the number of supernovae per century per 10¹⁰ L_⊙ galaxy far-infrared luminosity (see Section 2).

4

The definition of far-IR luminosity varies in the literature. Throughout this paper we adopt [W] (Sanders & Mirabel 1996).

5

All the distances given by Tully (1988) have been scaled to

6

The actual peak magnitude of core-collapse SNe is normally missed. ‘BV maximum light’ in this context usually means the brightest observed magnitude on the already declining light curve. However, in the case of SN 1983N, the B-band light curve was followed through the true maximum. In this paper we adopt the time of maximum light in the BV band as epoch 0 d.

7

iraf is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy Inc. under contract with the National Science Foundation.

Appendix A:

The galaxy sample for the discussed observing programme

Rowan-Robinson & Crawford (1989) considered the far-IR spectra of galaxies as arising from a mixture of (1) a cool disc component, (2) a warm starburst component, and (3) a hot Seyfert component. They determined the relative proportions, α_j, of the spectrum attributable to each of the components by fitting models to the far-IR spectra of the galaxies. They plotted IRAS colour—colour diagrams for the flux density ratios: versus and versus In these plots, the starburst galaxies occupy well-defined areas and thus the far-IR flux ratios can be used as selection criteria for the galaxy sample.

We selected galaxies from Soifer et al. (1987, 1989) and Rowan-Robinson & Crawford (1989), whose far-IR luminosity is greater than or comparable to those of M 82 and NGC 253, excluding galaxies whose far-IR luminosity is powered by a population of old stars or an AGN. We used the following criteria for selecting the sample.

• (1a)

  For galaxies with

• (1b)

  For galaxies with

• (1c)

  For galaxies with

• (2)

  Flux ratios and

• (3)

  From Rowan-Robinson & Crawford (1989) the galaxies with

• (4)

  Type 1 Seyfert galaxies listed in Veron-Cetty & Veron (1998) were excluded.

The flux ratios, far-IR luminosities (Sanders & Mirabel 1996) and distances (Tully 1988) for the selected galaxies are presented in Table A1, columns In columns a lower and upper limit for the intrinsic SN rate are presented according to (a) equation (4) with and (b) equation (5). In estimating the number of supernovae discoverable in an observing programme, we assumed that, for most of the 45 sample galaxies, the fraction of their far-IR luminosity that is powered by young massive stars is similar to that of NGC 253 and M 82. However, for the type 2 Seyfert galaxies in the sample we reduced this fraction by a factor of 2, with the SN rates correspondingly reduced.

Appendix B:

Infrared photometry of core collapse supernovae

We have assembled near-IR photometric data for a total of 14 SN events. SN 1987A has been excluded for the reasons outlined in the main text. The data are listed in Table B2.

Table B1.

Infrared photometry of core collapse supernovae.