A&A manuscript no.

(will be inserted by hand later)

Your thesaurus codes are:

missing; you have not inserted them

Direct Measurement of the Supernova Rate
in Starburst Galaxies

J.D. Bregman 1 , P. Temi 1 , and D. Rank 2

1 NASA Ames Research Center, Moffett Field, CA 94035
email: jbregman@mail.arc.nasa.gov

email: temi@ssal.arc.nasa.gov

2 University of California Santa Cruz, Santa Cruz, CA 95064
email: rank@ucolick.org

Received ; accepted

Abstract. Supernovae play a key role in the dynamics, structure, and chemical evolution
of galaxies. The massive stars that end their lives as supernovae live for short times.
Many are still associated with dusty star formation regions when they explode, making
them difficult to observe at visible wavelengths. In active star forming regions (galactic
nuclei and starburst regions), dust extintion is especially severe. Thus, determining the
supernova rate in the active star forming regions of galaxies, where the supernova rate
can be one or two orders of magnitude higher than the average, has proven to be difficult.
From observations of SN1987A, we know that the [Nill] 6.63 /xm emission line was the
strongest line in the infrared spectrum for a period of a year and a half after the explosion.
Since dust extintion is much less at 6.63 /mi than at visible wavelengths (A 6 . 6 3/Av =
0.025) , the Nill line can be used as a sensitive probe for the detection of recent supernovae.
We have observed a sample of starburst galaxies at 6.63 /im using ISOCAM to search for
the Nill emission line characteristic of recent supernovae. We did not detect any Nill line
emission brighter than a bo limit of 5 mJy. We can set upper limits to the supernova rate
in our sample, scaled to the rate in M82, of less than 0.3 per year at the 90% confidence
level using Bayesian methods. Assuming that a supernova would have a Nill with the
same luminosity as observed in SN1987A, we find less than 0.09 and 0.15 per year at the
50% and 67% confidence levels. These rates are somewhat less if a more normal type II
supernovae has a Nill line luminosity greater than the line in SN1987A.

Key words: Supernovae: general - galaxies: starburst

2 Bregman et Al.: SN rale in Starburst Galaxies

1. Introduction

While the supernova rate is one of the key parameters which constrain the ini-
tial mass function and rate of star formation in models of starburst galaxies, its
value is difficult to measure (Doane and Mathews 1993). Observational approaches have
been used to determine this rate, including observations of compact radio sources
(Kronberg et al. 1985, Antonucci and Ulvestad 1988, Van Buren and Greenhouse 1994)
and direct visible light imaging (Richmond et al. 1998). The radio observations suffer
from confusion since there are a large number of point sources, and there is an un-
certainty in the ages of the supernova remnants, leading to a range in the calculated
supernova rate of 0.08-0.3 supernovae per year in M82. The optical observations did not
detect any supernovae within the starburst regions of the observed galaxies, leading the
authors to conclude that there was too much obscuration within the galaxies' nuclei to
allow supernovae to be observed at visible wavelengths.

Van Buren and Norman (1989) suggested that supernovae in starburst galaxies could
be directly observed in the infrared by imaging the galaxies in the [Coll] 10.52/im emission
line (from radioactive cobalt) since it would be a unique signature of a supernova and
would provide information about the mass of the supernova. However, this line is exactly
coincident with the 10.52/im [SIV] emission line that is prevalent in planetary nebulae
and HII regions. The [Coll] line, from Co 56 , does decay with a half life of 77 days, so
it would be possible to distinguish the difference between [SIV] and [Coll] lines by their
temporal behavior if the [Coll] line was spatially separated from [SIV] line emission and
it was a substantial fraction of the energy at 10.52/im. Infrared spectra of SN1987A
(Rank et al. 1988, Wooden at al. 1993) showed the [Coll] line in emission, but a [Nill]
emission line, from Ni 58 . at 6.63/im was stronger and persisted longer than any other
line. The emission lines appear after the peak of the visible light curve when the envelope
becomes optically thin, about 120 days after the explosion. The [Nill] line was not evident
in a spectrum obtained 60 days after the explosion, but was already a strong line at the
next time of observation 260 days after the explosion. The line was weak but still visible
at the last observation 775 days after the explosion. Thus, the [Nill] line should be visible
in a supernova for nearly 2 years. This line is not seen in any other type of source, it is
well separated from any other emission line, and is a unique and long lived supernova
indicator which should be relatively free of obscuration from dust within the starburst
nucleus.

With the launch of the Infrared Space Observatory (ISO), we were provided the
opportunity to search nearby starburst galaxies for recent supernovae by imaging the
galaxies at 6.63 /im and at several comparison continuum wavelengths. We did not detect
any supernovae in the starburst nuclei, and to this date, no supernova has been detected

Bregman et Al.: SN rate in Starburst Galaxies
Table 1. Log of ISO Observations

.Xame

Obs. Date

Int. Time

Num. of frames

(sec.)

JG342

10/02/97

-

5.04

233

.VGC253

06/07/97

-

5.04

155

A/82

03/11/97

-

5.04

155

.VGG4449

07/07/96

-

10.08

76

.YGC4945

08/11/96

08/11/97

10.08

229

.VGC5128

08/11/97

-

5.04

155

JVGG5236

07/31/96

08/24/97

10.08

239

CIRCINUS

08/19/96

03/18/97

10.08

232

jVGG6946

10/20/96

04/30/97

10.08

228

A r GC5055

07/13/96

-

10.08

213

in a starburst nucleus. In this paper we derive a limit to the supernova rate in starburst
nuclei from observations of a sample of galaxies. The observations and data reduction
procedure are discussed in § 2, followed by a derivation of the supernova rate in § 3. Our
results are summarizes in § 4.

2. Observations and Data Analysis

Images of the galaxies listed in Table 1 were obtained with the infrared camera (ISO-
CAM) aboard ISO. We were fortunate enough to get two epochs of observations for some
of the galaxies, extending the coverage time by about a year. The first set of observations
were made using a single set of continuously variable filter (CVF) settings with spectral
resolution of A/AA ~ 45. In the second set of observations, we scanned the CVF from
short to long wavelengths, moved the telescope 1.5 pixels, then scanned the CVF in the
reverse direction. This provided us with redundant data, allowing us to better remove
artifacts from internal reflections in the camera. We also found that the longest integra-
tion time (20 seconds) produced images with so many cosmic ray trails that it was nearly
impossible to remove false signals. The same 10 CVF settings were used for all of the
observations, and cover the wavelength range from 6.04/xm to 7.795/im, including lines
from [Nill] (6.63/im), [Aril] (6.98jim) and [Pfa] (7.45/xm) and adjacent continuum. For
all of the observations we used the CAM04 AOT with 3 arcsec per pixel. A log of the
observations is given in Table 1.

Brcgman ct Al.: SN rate in Starburst Galaxies
M 82 NGC 253

NGC 4449

■e 4

R -Js

9|yii

' Hb9I

WmMi^

• s^^b

B^B«MP^m35_

i^Bt* 1 ^ s» ^"Si/^r^ir

- bHB

jfiBfe E alr ^^i^^&a^a.

^ " ifm^ a^fc-<mi^ -T^Jfct'

n*. **^U fywraRf

NGC 4945

IC342

NGC 5128

■ /?-

M

iSBi

9

*

NGC 6946

■

■

'■ #n

.c

o

' ■

o O

Fig. 1. Gray scale images of the galaxy sample at 6.63 /im overlaid with contour lines. Gray scale
images are shown with the original 3 arcsec/pixel scale, while the contours have been smoothed
with a 3 pixel FWHM gaussian filter. The lower contour in each galaxy is at a level equal to a
noise of 5 a.

2.1. Data Reduction

Data reduction has been performed using the Camera Interactive Analysis (CIA) package.
The new release uses the latest calibration files available and some new algorithms for
detector characterization and behavior. For the dark current subtraction, a dark model
has been used to take into account the long term-drift in the dark current which occurred

Bregman et Al.: SN rate in Starburst Galaxies 5

during the ISO mission. Great effort has been devoted to removing cosmic ray hits on
the detector. In order to detect a supernova event, we needed to generate continuum
subtracted images of our target galaxies at 6.63 fxm. The galaxy's nucleus itself is a fairly
bright point-like source, but the expected supernova signature is of the order of a few
to tens of mJy. Thus, a clear detection on the subtracted frames relies heavily on the
removal of cosmic rays and correction for the glitch-induced transient responses. Most of
the short term duration glitches can be easily removed using a sigma clipping filter or
the muli-resolution method. These glitches are produced by protons and electrons hitting
the detector and their effects on the data are the typical spikes present on a pixel time
history plot. The spike duration is shorter than the integration time and they last for no
more than one or two readouts. Highly energetic protons and electrons and heavy ions are
responsible for glitches that cannot be easily removed. These special glitches introduce
a gain variation with a very long time constant. The gain stabilization can be very slow,
and at the present time there is no method that corrects the effect in a satisfactory way.
However, since we had redundancy in our data, we were able to check for remnants of
long term response variations and either remove or mask these occurrences on the original
CVF scan data frames.

The standard calibration file cal-g that comes along with the ISO data products is not
very well suited for flat field correction with data taken with the CAM04 AOT. We built
our own flat field combining calibration files that match our specific data set. After dark
current correction, deglitching and stabilization, we can represent a set of data taken
with the camera in the same configuration c, imaging a source on the array, as

I obs {x,y,c) = F(Source(x,y) + Zodiacal (x,y)) x dflat(x,y) x oflat{x,y) +

F(Scatt(x,y)) x dflat{x,y) (1)

The library flat field is separated in two distinct frames: the optical flat and the
detector flat. Here, Zodiacal{x,y),oflat{x,y) and dflat{x,y) correspond to the Zodiacal
light, the optical flat and the detector flat respectively as a function of pixel position (x,y).
Scatt(x,y) is the scattered light pattern which is produced inside the camera between
detector and filter and is thus unaffected by the oflat(x,y). We can re-write the same
relation as

I b,(x,y,c) = F{Source(x,y) x oflat(x,y) x dflat(x,y)) +

dflat{x,y) x {F{Zodiacal{x,y)) x oflat(x,y) + F{Scatt(x,y))} (2)

If the source does not produce too much scattered light, F(Scatt{x, y)) can be ignored
and the second term in equation (2) can be measured by scanning the zodiacal light.
A good quality data cube for the zodiacal light, covering the entire CVF wavelength
range, is not yet available. Using library CVF flat images, we built, by interpolation, a

Brcginan ut Al.: SN rate in Starburst Galaxies

15 20
Pixel

Fig. 2. 6.63 /im continuum subtracted image of IC 342. Three row plots taken in the central
region of the image show the peak to peak noise of the order of 3 mjy. A residual cosmic ray
strip is seen in the lower right corner of the image.

datacube that matches the wavelengths in our measurements. Scaling this cube to the
intensity measured off-source in our observations, and subtracting the scaled cube from
the original data set, we are left with the F(Source(x, y)) x oflat(x, y) x dflat(x, y) term.
The optical flat shows little structure in the central part of the array where our sources
are centered. Thus there is no need to correct for it. This is not the case for the detector
flat. We recovered some detector flats from the narrow band filter calibration files, and
re-sampled them to match the wavelengths in our CVF scan. This way we created a
detector flat cube we could use to flat field our data.

Figure 1 shows a panel of our galaxy sample. For each set of CVF scans we have
produced 6.63pm continuum subtracted images; the \a rms noise level on these maps is
of the order of 0.8 mjy per pixel. Figure 2 displays a subtracted frame for IC 342 as an
example of a fully reduced image. Low frequency noise from glitches is still the major
contributor to the overall noise in the figure. If a possible detection were present in one
frame, we had the opportunity to confirm the detection using the data taken with the
CVF scanned in the opposite direction. We did not detect any clear emission from a
supernova explosion in our sample using this criterium.

Bregman et Al.: SN rate in Starburst Galaxies 7

3. Discussion

3.1. Comparison of [Nil I] luminosity in SN1987A to a normal supernova

SN1987A is the only supernova bright enough to have been observed in the mid-infrared,
and our knowledge of the [Nill] luminosity in supernovae is based soley on this one object.
We therefore need to know how SN1987A compares to other Type II supernovae before
we can determine the expected brightness of a supernova produced [Nill] line in other
galaxies.

The radioactive tail of a supernova light curve is produced by thermalization of
gamma rays from Co 56 decay. The Co 56 itself comes from rapid decay of the Ni 56
produced in the supernova explosion. In most supernovae, almost all of the gamma rays
are absorbed in the expanding envelope, so that the luminosity on the tail is a good
indication of the amount of Ni 56 produced by the supernova. Even though SN1987A
did not achieve the peak luminosity of most Type II supernovae, it did have about
the same luminosity as typical Type lis after its peak (eg. Turatto et al. (1998)), and
therefore had produced about the same amount of Ni 56 as a typical Type II supernova,
as expected from theoretical calculation of Ni 56 production (see Figure 3).

The [Nill] 6.63 /im line observed in SN1987A is a collisionally excited ground state
forbidden line of Ni 58 , which is also produced in the supernova explosion, and, like the
bolometric luminosity, is powered by radioactive decay of Co 56 . Gamma rays from the
cobalt ionize the gas in the supernova envelope, heating the electrons. The [Nill] line
intensity therefore depends on the density, on the quantity of Ni 56 and Ni 58 produced
in the explosion, on the fraction of Ni and Ni + , and on the temperature of the gas. The
[Nill] line becomes visible when the optical depth in the expanding envelope becomes
optically thin above the Nill zone, and is coincident with the start of the tail phase of a
supernova. A typical Type II supernova takes 40-60 days to reach this phase, while the
much slower expanding SN1987A took 120 days. However, the combined slower expansion
and longer time results in the density of the Nill zone being about the same in SN1987A
as in a typical type II. Also, due the high density of the Nill emitting region, the [Nill]
6.63 jzm line intensity is only linearly dependent on the density. The amount of Ni 56 and
Ni 58 produced as a function of mass and metallicity have been calculated by Woosley and
Weaver (1995), and are plotted in Figure 3 for both solar and 0.1 solar metallicity. The
LMC is about one third solar, so we would expect the Ni production to lie between these
two curves. For comparison, 0.075 Mq of Ni 56 was produced by SN1987A, which agrees
quite well with Woosley and Weaver calculations. A good indicator of the trend of the
[Nill] emission line intensity is the product of Ni 56 and Ni 58 , which is shown in Figure 4.
A supernova like SN1987A is near the minimum of the [Nill] line emission as a function

Bregman et Al.: SN rat- .:. • .,rburst Galaxies

1.000

. 0.100

o 0.010

o

en
o

0.001

Ni 56 (z=0.1 SN1987A

Ni" (z=1)

Ni 58 (z=1)

x-

\

(z = 0.1)

\

\

10

25

15 20

Stellar Mass (M Q )

Fig. 3. The masses of Ni 56 and Ni 58 produced in a supernova explosion as calculated by Woosley
and Weaver (1995) are shown as a function of stellar mass for both solar and 0.1 solar metallicity.
Ni 5e rapidly decays to Co 56 which then produces the gamma rays that power the supernova
during the tail phase when forbidden lines are visible. Ni 58 is stable and is the source of the
[Nill] 6.63 /tm emission line.

of mass, and significantly below that expected for a solar metallicity precursor. Thus, it
is likely that the supernovae in galaxies with solar metallicities could have Nill emission
lines from several times to an order of magnitude more luminous than that observed
in SN1987A. In our [Nill] emission line estimates, we will use both the SN1987A value
(6 x 10~ 17 Wcm~ 2 Wooden at al. 1993) and a value 3 times greater than that observed
in SN1987A. However, we should keep in mind that this value is very uncertain and we
would learn a lot about Nill forbidden line emission if we observe just one supernova.

3.2. Visibility time for the sample

The measured [Nill] line intensity in SN1987A at four different epochs after the explosion
(260, 415, 615 and 775 days) are reported by Wooden et al., (1993). The line intensity
reaches its peak at 415 days after core collapse with an intensity of 6.15 x 10 -17 Wcm~ 2 .
At 775 days after collapse, the line intensity drops to 4.37 x 10 -18 Wcm~ 2 , but is still the
brightest emission line at infrared wavelengths. From these measurements we calculated
the expected luminosity of a supernova occurring in each galaxy of the sample. The
estimated peak signal that a supernova as bright as SN1987A would produce (in each

Bregman et Al.: SN rate in Starburst Galaxies

0.0100

» X 0.001 Or

o

if)
o

0.0001 -

25

15 20

Stellar Mass (M Q )

Fig. 4. The product of Ni 56 and Ni 58 is shown as function of stellar mass of a supernova progen-
itor for solar and 0.1 solar metallicity stars. Since Ni 56 provides the gamma rays to power the
supernova and the 6.63 /*m line is from Ni 58 , their product is an indicator of how the [Nill] 6.63
/im line should vary with stellar mass and metallicity. The average solar metallicity supernova
has a Ni 56 x Ni 58 product 3-4 times that of SN1987A.

sample galaxy), is shown in figure 5. Since the noise level in the continuum subtracted
maps is of the order of 0.8 mJy, we define a visibility time, t vU , as the period of time
in which the signal produced at 6.63 ^m by the [Nill] line would be > 5 mJy, in each
galaxy in our sample. This will allow us, in case of an event, to make a clear detection
with a S/N ratio > 5. The visibility time in the sample ranges from ~ 1 year to more
than 1.5 years for the closest galaxies; for a supernova with a Nill line emission 3 times
brighter than SN1987A the visibility time for the sample increases by about 20%. For
NGC 5055 the estimated signal from the Nill line, assuming a supernova as bright as
SN1987A, never reaches the 5 mJy level required for a clear supernova detection with our
data, while it has a visibility time of 1.2 years assuming a 3 times brighter supernova.

To take into account the fact that some of the galaxies have been observed twice
during the ISO mission we need to compute the control time, C" ime , reference for each
galaxy of the sample defined as:

3

time

= $>«

(3)

i=l

10

Bregman ct Al.: SX rate in Starburst Galaxies

x

3

800

where:

At< =

(4)

iV« if ti -U-i > t?" or i = l

U - <i_! if U - ti-! < tf* and i^\

Here U is the epoch of the i tf> observation of the j th galaxy. For those galaxies that
have been observed only once, the control time C t>me coincides with the visibility time
t v ". Control times are reported in column 5 of Table 2.

3.3. Derivation of the rate

In order to derive the supernova rate for our galaxy sample, we need some way of scaling
the rates for the individual galaxies. Since the starburst nuclei are dusty, most of the
energy produced by stars in the starburst will be absorbed by the dust and re-radiated
at far infrared wavelengths. Thus, we can use the far infrared luminosities of the galaxies
based on IRAS fluxes to scale the expected supernova rate for each galaxy (eg. Soifer et
al. (1989)). This scaling will only be valid if the galaxy nuclei are optically thick in the
visible, and if the FIR luminosity measured by IRAS is confined to the region we observe.
For starburst galaxies, the starburst nucleus has a radius of a few hundred parsecs (Doane
and Mathews, 1993). For M82, a 400 pc diameter starburst region extends for 25 arcsec,
while our ISOCAM images extend for 90 arcsec. The galaxy images also show that the
infrared emission is strongly concentrated near the nucleus.

Bregman et Al.: SN rate in Starburst Galaxies

11

50
40

-^ 30

x

.5 20

LL

10r

— i 1 r-

Ic342

(b)

200

400
Days

600

800

Fig. 5. Visibility time for the galaxy sample. The 5 a limit for a supernova detection is showen
as a dotted line. The expected Nill emission line intensities as a function of time after the
explosion have been scaled from the measured intensity of SN1987A at different epochs.

From our data, we can derive a confidence level as a function of the supernova rate
using Bayesian probabilities (see Sivia, 1996, for an excellent discussion of Bayesian data
analysis). Bayes theorem is appropriate for data analysis where we have data and want
to derive the probability of a model being true. In this case, the model is simply that
supernovae occur randomly with a rate that gives the average time between supernovae
which can be detected for a time At by observing the [Nill] emission line. The basic
equation from Bayes theorem is usually stated as

prob(M\D,I) = prob(D\M,I) x prob(M\I)

(5)

where M = the model, D = the data, and I = prior information, and all of the prob( ) are
probability density functions. The vertical bar means given, so the first term reads the
probability of the model given the data. As with all probabilities, the total probability is
one. The last term is the probability of the model being true given our knowledge of the
model and the way our observing process occurs. Since we observe the galaxies at discreet
times, t obs , then we will only observe a supernova if it has occurred within a short time,
At, before the observation. There are then two variables in our model, the supernova
rate, which we will call A, and the time during which a supernova could be detected, At.
The probability of the data being true, which is that no supernova was seen, is 1 if the

12 Bregman et Al.: SN rate in Starburst Galaxies

Table 2. Control Times

Name

d a

Lfir

SN Peak

Control Time

Control Time(3x) h

(Mpc)

(10 10 L Q )

(mJy)

(Days)

(Days)

7C342

2.1

0.23

43.62

667

759

7VGC253

2.5

1.09

30.78

621

759

M82

3.3

1.98

17.66

529

705

JVGC4449

3.7

0.08

14.05

483

667

NGC4945

3.9

1.75

12.64

825

1017

NGC5128

3.9

0.64

12.64

460

652

NGC5236

3.9

0.79

12.64

849

1041

CIRCINUS

4.0

1.20

12.02

656

855

NGC6946

5.5

0.83

6.36

406

763

NGC50S5

7.2

0.53

3.71

429

* Distances for each galaxy in the sample are taken from the following authors:
(Karachentsev et al.,1993)(IC 342), (Sreekumar et al., 1994)(NGC 253)
(Doane and Mathews 1993)(M82), (Bajaja et al., 1994)(NGC 4449)

(Bergman et al., 1992)(NGC 4945, NGC 5128, NGC 5236), (Curran et al., 1998)(Circinus)
(Tully, 1988)(NGC 6946, NGC 5055), (Doane and Mathews 1993)(NGC 7714).

b Control Time based on a supernova with a Nill emission line 3 times brighter than SN11987A

observation occurred outside of the observing window and if the observation occurred
within At of the explosion. If the explosion occurs at ti, then

prob(M\I)

if t 1 + At < t 0b3

if h < Ubs < h + At

(6)

We are then left with determining the prob(D | M,I), or prob(D | A, At), which is the
distribution of intervals between events which have a Poisson distribution, and is given
by

P(\,t) = Ae ( - A "

(7)

We are really interested in this probability integrated over time, keeping in mind that
it is multiplied by prob(M | I). Recalling that the total probability must equal 1, the
normalized probability density function of interest is therefore

prob(M\D,I) = At e { ~ XAt)

(8)

Our goal is to place an upper limit on the supernova rate from our data, and thus we
should integrate this probability density function over all rates less than a maximum

Bregman et Al.: SN rate in Starburst Galaxies

13

&S

(L>
O
C
CD

c
o
o

100

r i

■■■'■ ■ i ■ '

T 1— r - T 1 . . | . . . . | . •

80

-

SN 3x brighter

-

60

''SN as SN1987A

-

40

-

20

-

. . i . . , , i , ,

. . i . . , . i , . , , i . ,

. . i ,

0.00 0.05

0.10 0.15 0.20 0.25
Supernova Rate

0.30

Fig. 6. The confidence (probability) of seeing a supernova in the galaxy sample as a funcion of
the supernova rate in M82. The solid curve is for a supernova with a Nill line bright as SN1987A,
while the dotted line assumes 3x brighter Nill line.

rate. Integrating over rates from zero to A m gives, for a single galaxy, the probability
that the rate is less than A m of

J°(A < A m ) = 1 - e { ~ XmAt)

(9)

For the complete sample of galaxies, we multiply the probabilities of not seeing a super-
nova in each galaxy when the rate is less than A m , then subtract that value from 1 to
give the probability of seeing a supernova in our sample. Figure 6 shows the confidence
(or probability) of seeing a supernova as a function of the maximum supernova rate
for the entire galaxy sample scaled to M82 for both the SN1987A control time and the
3x SN1987A control time. The 50% confidence rate, that is the rate which has an equal
probability of being correct or incorrect, is 0.09/0.065 supernovae per year in M82, where
the first listed rate is for supernovae as bright as SN1987A. The 67% rate is 0.15/0.11
supernovae per year, while there is a 90% confidence that the rate is less than 0.30/0.23
supernovae per year in M82.

Supernova rates calculated by other authors are usually based on an approch which
assumes a constant supernova rate. In that case, the probability of seeing a supernova
is just the control time divided by the rate. This approch gives rates at fixed confidence
levels which are somewhat less than we calculated using a Bayesian method, resulting

1-1 Bregman et Al.: SN rate in Starburst Galaxies

in rates of 0.09/0.065, 0.135/0.10. and 0.25/0.19 for confidence levels at 50%, 67%, and
90% respectively.

By comparison, Van Buren and Greenhouse (1994) derive a rate of 0.1 supernovae per
year from radio source counts, assuming that all observed radio sources are supernova
remnants, and modeling the evolution of the source brightness with time. They used
data from Kronberg et al. (1985), who had derived a rate about twice their value, and
quote a result of 0.2 per year from the evolution of radio sources observed by Kronberg
and Sramek (1985). Doane and Mathews (1993) and Rieke et al. (1993) considered a
supernova rate for M82 of between 0.1-0.3 as fitting the observational data. It is clear
from our direct search for recent supernovae that the rate in M82 is highly unlikely to
be greater than 0.2 per year, and it is doubtful that there are many supernovae hidden
within dusty starburst nuclei or that there is any enhanced star formation in the galaxy
cores.

4. Summary

We have observed a sample of starburst galaxies at 6.63 fim using ISOCAM to search
for the Nill emission line characteristic of recent supernovae. While we did not detect
any supernovae, we can put upper limits on the supernova rate. For example, the rate in
M82 is less than 0.3 supernovae per year with a confidence of 90% assuming that the Nill
line in a supernova will have the same luminosity as the line observed in SN1987A. For a
supernova with a Nill line 3 times brighter than in SN1987A, the rate is less than 0.23 per
year at the 90% confidence level. The corresponding rates for 50% and 67% confidence
levels are 0.09/0.065 and 0.15/0.11 supernovae per year. There is thus no evidence in our
data for an enhanced supernovae rate in the nuclei of starburst galaxies.

References

Allen, M. L., and Kronberg, P. P. 1998, ApJ, 502, 218.

Antonucci, R. R. J., and Ulvestad, J. S 1988, ApJ, 330, L97.

Bajaja, E., Huchtmeier, W.K., Klein, U., 1994, A&A, 285,385.

Bergman, P., Aalto, S., Black, J. H., Rydbeck, G., 1992, A&A, 265, 403

Cappellaro, E., Turatto, M., Tsvetkov, D., Bartunov, O. S., Pollas, C., Evans, R., Hamuy, M.,

1997, A&A, 322, 431.
Curran, S. J., Johansson, L. E. B., Rydbeck, G., Booth, R. S., 1998, A&A, 338, 863
Doane, J. S., and Mathews, W. G 1993, ApJ, 419, 573.

Karachentsev, I. D.; Tikhonov, N., 1993, A&AS, 100,227. Kronberg, P. P., Biermann, P., and

Schwab, F. R. 1985, ApJ, 291, 693.
Kronberg, P. P., and Sramek, R. A. 1985, Science, 227, 28.

Bregman et Al.: SN rate in Starburst Galaxies 15

Rank, D. M., Pinto, P. A., Woosley, S. E., Bregman, J. D., Witteborn, F., Axelrod, T. S., and

Co hen, M. 1988, Nature, 311, 505.
Rice, W., Lonsdale, C. J., Soifer, B. T., Neugebauer, G., Kopan, E. L., LLoyd, L. A., De Jong,

T. and Habing, H. J., 1988, AJ, 68, 91
Richmond, M. W., Filippenko, A. V., and Galisky, J. 1998, PASP, 110, 553.
Rieke, G. H., Loken, K., Rieke, M. J., and Tamblyn, P. 1993, ApJ, 412, 99.
Sivia, A., A Bayesian Tutorial, Claredon Press, Oxford 1996.

Soifer, B. T., Boehmer, G., Neugebauer, G., and Sanders, D. B., 1989, AJ, 98, 776
Sreekumar, P., Bertsch, D. L., Dingus, B. L., Esposito, J. A., Fichtel, C. E., Hartman, R. C.,

Hunter, S. D., Kanbach, G., Kniffen, D. A., Lin, Y. C, Mattox, J, R., Mayer-Hasselwander,

H. A., Michelson, P. F., Von Montigny, C., Nolan, P. L., Schneid, E. J., Thompson, D. J.,

1993, ApJ, 426,105
Tully, R.B., 1988, Nearby Galaxies Catalog.
Turatto, M., Mazzali, P. A., Young, T. R., Nomoto, K, Iwamoto, K., Benetti, S., Cappellaro, E.,

Danziger, I. J., De Mello, D. F., Phillips, M. M., Suntzeff, N. B., Clocchiatti, A., Piemonte,

A, Leinbndgut, B., Covarrubias, R., Maza, J., Sollerman, J., 1998, ApJ, 498, L129.
Van Buren, D., and Greenhouse, M. A. 1994, ApJ, 431, 640.
Van Buren, D., and Norman, C. A. 1989, ApJ, 336, L67.
Wooden, D. H., Rank, D. M., Bregman, J. D., Witteborn, F. C, Tielens, A. G. G. M., Cohen,

M., P into, P. A., and Axelrod, T. S. 1993, ApJS, 88, 477.
Woosley, S. E., and Weaver, T. A. 1995, ApJS, 101, 181.