A Survey of Novae in M83

All images were pipeline processed in the usual manner by first subtracting the bias and dark current, and then flat-fielding the individual images to remove the high-frequency, pixel-to-pixel variations using APHOT (a synthetic aperture photometry and astrometry software developed by M. Velen and P. Pravec at the Ondřejov observatory; Pravec et al. 1994). To eliminate cosmic-ray artifacts in our nightly images, we obtained a series of 120 s images that were later spatially registered and median stacked using SIPS ¹³ to produce a final image for a given night of observation intended for nova searching. Photometric and astrometric measurements of the novae were done using APHOT on spatially registered and stacked nightly images.

Instrumental magnitudes for all nova candidates were determined by summing the fluxes in 2'' diameter circular apertures in all epochs in which they were visible. Calibrated R-band magnitudes were then determined by differential photometry with respect to a set of six secondary standard stars in the M83 field. Both the R-band and I-band magnitudes of the secondary standard stars were photometrically calibrated by us using the same instrumentation as we used for the survey. As primary standards we used the stars and their magnitudes published in Landolt (1992). We observed both the primary standards and the M83 field during one night under excellent photometric conditions to properly calibrate the secondary standards in the field of M83, which allow us to obtain photometry of the novae also in nonphotometric nights.

The calibrated magnitudes of all 19 novae discovered as part of our survey, on all nights where they were visible, are presented in Table 3. The temporal sampling of our survey was sufficient to produce useful R-band light curves for all but one of the 19 novae. The light curves for these 18 novae having multiple epochs of observation are shown in Figure 4. The properties of the light curves (i.e., their peak magnitudes and fade rates) will be explored further in Section 6.

Zoom In Zoom Out Reset image size

A determination of the limiting magnitude of our survey images is complicated by the fact that the galaxy background surface brightness is highly variable and that the spatial distribution of the novae cannot be assumed to be uniform across the survey images. Thus, no one limiting magnitude can represent the coverage of a given image. We have approached this problem following the procedure described in our earlier work on NGC 2403 (Franck et al. 2012). Specifically, we have conducted artificial nova tests on a representative image (hereafter the fiducial image) under the assumption that the spatial distribution of the artificial novae follows the background light of the galaxy.

The artificial novae were generated using tasks in the IRAF DAOPHOT package, which enabled us to match the point-spread functions of the real stars in the image. Using the routine addstar, the fiducial image was then seeded with 100 artificial novae having apparent magnitudes randomly distributed within each of a total of eight, 0.5 mag wide, bins. For each of the magnitude bins, the artificial novae were distributed randomly, but with a spatial density that was constrained to follow the integrated background galaxy light of M83. We then searched for the artificial novae using the same procedures that we employed in identifying the real novae. The completeness at the fainter magnitudes was somewhat higher using the ISIS image subtraction analysis, but was generally consistent with the results from a direct comparison of median-subtracted images.

The fraction of novae recovered from the two search techniques in each magnitude bin yielded the completeness functions shown in Figure 7. Given that we discovered the same number of novae in M83 employing both techniques, we have chosen to take the average completeness in a given magnitude bin (the heavy line in Figure 7) to form the basis of the completeness function, C(m). To allow for uncertainty in magnitude bins where the completeness functions differ (the shaded regions), we randomly sample allowed values of C(m) in the analysis to follow. This completeness function can be generalized to any epoch, i, of observation by applying a shift, Δm_i ($={m}_{\mathrm{lim},0}-{m}_{\mathrm{lim},i}$), which represents the difference in the limiting magnitudes (as measured by the faintest star that could be reliably detected near the perimeter of the image away from the galaxy) of our fiducial image and that of the ith epoch image. Thus, for any epoch, i, we have C_i (m) = C(m + Δm_i ).

Zoom In Zoom Out Reset image size

As in our earlier extragalactic nova studies (e.g., Franck et al.2012; Güth et al. 2010; Coelho et al. 2008; Williams & Shafter 2004), we have employed a Monte Carlo simulation to compute the number of M83 novae that we would expect to observe during the course of our survey.

For a given assumed annual nova rate, R, we begin by producing trial novae erupting at random times throughout the time span covered by our survey, each having a peak luminosity and fade rate that has been selected at random from a large sample of known R-band light-curve parameters. Ideally, we would like to use light-curve parameters specific to the full population of M83 novae, but such an unbiased sample does not exist. Instead, we have used the M83 light-curve parameters from our observations given in Table 5, augmented with additional R-band light-curve parameters from the M31 light curves observed by Shafter et al. (2011).

Adopting a distance modulus to M83 of μ₀ = 28.34 ± 0.14, which represents the mean of the Cepheid and the tip of the red giant branch distances from the recent study by Tully et al. (2016), enables us to compute the expected apparent magnitude distribution at any given epoch, i, during our survey, n_i (m, R). To account for uncertainty in the distance, our numerical simulations also randomly select values of the distance modulus normally distributed about the mean value. The number of novae expected to be detectable during the course of our survey, N_obs(R), can then be computed by convolving the simulated apparent magnitude distribution with the completeness function, C_i (m), and then summing over all epochs of observation:

$$\begin{eqnarray}&&{N}_{\mathrm{obs}}(R)=\sum _{i}\sum _{m}{C}_{i}(m){n}_{i}(m,R).\end{eqnarray} \tag{ 1 }$$

The intrinsic nova rate in M83, R, and its uncertainty can now be determined through a comparison of the number of novae found in our survey, n_obs = 19, with the number of novae predicted by Equation (1). We explored trial nova rates ranging from R = 1 to R = 50 novae per year, repeating the numerical simulation 10⁵ times for each trial value of R. The number of matches, M(R), between the predicted number of observable novae, N_obs(R), and the actual number of novae discovered in our survey, n_obs = 19, was recorded for each trial value of R. The number of matches was then normalized by the total number of matches for all R to give the probability distribution function, P(R) = M(R)/∑_R M(R) shown in Figure 8. The most probable nova rate in the portion of the galaxy covered by our survey images is ${18}_{-3}^{+5}$ yr⁻¹, where the error range (1σ) for the asymmetrical probability distribution has been computed assuming it can be approximated by a bi-Gaussian function.

Zoom In Zoom Out Reset image size

The K-band photometry of M83 from the Two-Micron All Sky Survey (2MASS) suggests that the extended halo may extend out to a distance of ${r}_{\mathrm{tot}}\sim 8\buildrel{\,\prime}\over{.} 5$ from the center of M83. Thus, it is possible that our survey images may be missing a small fraction (≲5%) of the light from the extended halo of M83. Under the assumption that we are sampling 0.95 ± 0.05 of the total M83 light, we estimate the global nova rate for M83 to be ${R}_{M83}={19}_{-3}^{+5}$ yr⁻¹.

In order to compare the nova rates between different galaxies or different stellar populations, the rates must first be suitably normalized. Ideally, it would be appropriate to normalize the rates by the mass of stars in the region surveyed, but the mass cannot be measured directly. As a proxy for the mass in stars, it has become standard practice to normalize the nova rate by the infrared K-band luminosity of the galaxy. In the case of M83, the integrated apparent K-band magnitude as measured by 2MASS is K = 4.62 ± 0.03. Given the distance modulus, μ₀ = 28.34 ± 0.14, and taking the absolute K-band magnitude of the Sun to be M_K = 3.27 ± 0.02 (Willmer 2018), we find that M83 has an absolute magnitude in the K band of M_K =− 23.72, and a corresponding K-band luminosity of (6.32 ±0.84) × 10¹⁰ L_⊙,K . Since we estimate that our survey covers 0.95 ± 0.05 of the entire galaxy where we have found an overall nova rate of ${18}_{-3}^{+5}$ yr⁻¹, we arrive at a K-band luminosity-specific nova rate for M83 of ${\nu }_{K}=({3.0}_{-0.6}^{+0.9})\times {10}^{-10}$ yr⁻¹ ${L}{}_{K,\odot }^{-1}$.

As recently reviewed by Shafter (2019) and Della Valle & Izzo (2020) prior to the present study, luminosity-specific nova rates had been measured for a total of 15 external galaxies. Figure 9 shows our value of ν_K for M83, along with the values for the other 15 galaxies taken from Table 7.3 in Shafter (2019), plotted as a function of the B − K color of the host galaxy. The specific nova rate for M83 is consistent with those of other spiral galaxies with measured nova rates. As we have noted in previous studies, despite the relatively high ν_K values reported for the Magellanic Clouds and M87—galaxies with very different Hubble types—there is no compelling evidence that ν_K varies systematically with the underlying stellar population.

Zoom In Zoom Out Reset image size

The MMRD relation for novae was first introduced by Mclaughlin (1945), who discovered that the most luminous members of a sample of 30 (mostly Galactic) novae faded more quickly than did their fainter counterparts. ¹⁴ He was able to quantify a linear MMRD relation of the form, $M=a\,+b\,\mathrm{log}({t}_{3})$, where a and b are fitting parameters and t₃ is the time it takes a nova to fade 3 mag from maximum light (in recent years t₂, which is more easily measured, is often used as an alternative). Over the years the MMRD relation has been calibrated many times, both in the Galaxy (e.g., Cohen 1985; Downes & Duerbeck 2000), and in nearby galaxies such as M31 (e.g., Capaccioli et al. 1989; Shafter et al. 2011), and has often been used as a means for determining the distances to novae where the apparent magnitude at maximum light and the fade rate have been measured.

Over the years it has become increasingly apparent that there is significant scatter in the MMRD, with the existence of the relation itself being called into question, initially by (e.g., Kasliwal et al. 2011) who found that a number of M31 novae observed with the Palomar Transient Factory (PTF) appeared to fall below the canonical MMRD relation (but, see Della Valle & Izzo 2020 for a different interpretation). Recently, it has become clear that a small subset of novae, typically those with massive white dwarfs that are accreting at high rates (e.g., recurrent novae) fade rapidly despite having relatively low peak luminosities, and thus deviate sharply from the MMRD relation. A good example is the M31 recurrent nova M31N 2008-12a (Darnley et al. 2014; Tang et al. 2014). Given that such systems do in fact exist, it is clear that not all novae will follow a universal MMRD relation. On this basis it has been recently suggested that the notion of an MMRD relation should be abandoned altogether (Schaefer 2018). However, the question of whether it is useful to continue to refer to an MMRD relation would seem to depend on the relative frequency of outliers. If the so-called "faint and fast" (FFN) novae, such as M31N 2008-12a, are intrinsically rare, then continuing to refer to an MMRD might make sense when considering the behavior of the majority of (nonrecurrent) novae. On the other hand, if such systems are relatively common, but just missed in most surveys that lack the depth and cadence to discover them, then perhaps the existence of an MMRD relation would be best considered as resulting from an observational selection effect. In either case, it appears that broadly speaking, luminous novae do on average fade more quickly than do their low luminosity counterparts.

Figure 10 shows the MMRD relation for the 17 novae in M83 where the light curves were sufficiently complete to allow a measurement of the peak magnitude and the rate of decline (here measured as t₂, the time for the nova to fade 2 mag from peak). Although there is significant scatter as expected, there is no question that a general trend, where the more luminous novae fade more quickly, is apparent. The best-fitting linear MMRD relation is given by: ${{\rm{M}}}_{R}=(-10.79\pm 0.42)+(1.96\pm 0.29)\,\mathrm{log}\,{t}_{2}$. The M31 R-band MMRD relation from the study of Shafter et al. (2011) is shown for comparison, and is remarkably similar. ¹⁵ Whether or not we are missing a putative population of FFN novae in M83 is unknown. The question can only be answered by future studies having greater depth and cadence, such as those that will be possible with the Large Synoptic Survey Telescope (LSST; Ivezić et al. 2019).

Zoom In Zoom Out Reset image size

As part of a comprehensive spectroscopic and photometric survey of novae in M31, Shafter et al. (2011) determined R-band light-curve parameters for a total of 42 novae and found that, as in the case of M83, the M31 nova sample also generally followed an MMRD relation, albeit with significant scatter (see Table 6 and Figure 19 in Shafter et al. 2011). The M31 novae are characterized by a mean absolute magnitude, 〈M_R 〉 = −7.55 ±0.14 and $\langle {t}_{2}(R)\rangle ={38}_{-5}^{+6}$ day, while their M83 counterparts from Table 5 yield 〈M_R 〉 = − 8.06 ± 0.21 and $\langle {t}_{2}(R)\rangle ={25}_{-5}^{+6}$ day. ¹⁶

It is interesting to compare the cumulative distributions of the peak luminosities and fade rates from each galaxy, as shown in Figures 11 and 12. The results of K-S tests show that the peak luminosity and fade rate distributions differ between each galaxy with 94% and 74% confidence, respectively. Thus, there is some evidence that the novae in the M83 sample are, on average, more luminous at maximum light and perhaps fade somewhat more rapidly compared with the M31 sample. A possible explanation for this difference is that M31 is an earlier Hubble type galaxy, SA(s)b, and the nova population observed by Shafter et al. (2011) was predominately a bulge population. On the other hand, as discussed earlier, our M83 sample appears to be primarily associated with the galaxy's disk. Taken together, Figures 11 and 12 provide additional support for the existence of two populations of novae, as originally suggested for Galactic systems three decades ago by Duerbeck (1990) and Della Valle & Livio (1998) based on photometric and spectroscopic observations, respectively.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The determination of extragalactic nova rates is challenging, with many sources of uncertainty that must be properly considered in the analysis. Among these, uncertainty in the determination of the survey completeness is perhaps the most important. As described earlier, the completeness—the fraction of novae erupting in the galaxy that can be detected in a given survey—depends quite sensitively on the intrinsic properties of the novae and their actual distribution within the galaxy under investigation. Since neither of these factors are known a priori, assumptions concerning both must always be made.

To guard against potential bias in our M83 nova sample, we chose to include light-curve parameters from previous observations of novae in M31 in our Monte Carlo nova rate simulations. As a check on the sensitivity of the final nova rate to the inclusion of the M31 light-curve parameters, we have also performed the analysis using only the light-curve parameters from the M83 novae discovered in our survey. Despite the differences in the light-curve properties between the two samples discussed in the previous section, the final nova rate determination is essentially unchanged (the rate ranges from 17 yr⁻¹ to 20 yr⁻¹ when the analysis is restricted to the M83 and M31 light-curve parameters, respectively). In this case, the insensitivity of the nova rate to the choice of light-curve parameters results from the fact that the M83 novae generally follow an MMRD relation. The M83 novae are on average brighter compared with their M31 counterparts, and thus easier to detect in our simulations, but they fade generally more quickly, so they are observable on average for less time. Thus, the two effects tend to cancel, leaving the final nova rate insensitive to the choice of light-curve parameters.

The form of the nova spatial distribution that we have adopted in the artificial nova completeness experiments can also affect the final nova rate computation. In the absence of strong evidence to the contrary, the default approach—and the one we have followed here—is to assume that the spatial density of novae follows the surface brightness of the galaxy (i.e., more light, more novae). However, in the case of M83, it appears that this assumption may be violated. As shown in Figure 5 the observed M83 nova spatial distribution is significantly more extended than the galaxy's R-band light. If this result accurately reflects the intrinsic nova spatial distribution (i.e., we are not missing novae in the inner regions of the galaxy), our artificial nova simulations may underestimate the overall completeness of our survey by placing a greater fraction of the artificial novae in the bright central regions of the galaxy where they are generally more difficult to detect. This bias, although we consider it to be small, would have the effect of slightly overestimating the nova rate determined earlier in our Monte Carlo simulations.

On the other hand, it is worth noting that M83 is an active star-forming galaxy (e.g., see Calzetti et al. 1999), which likely interacted ∼1–2 GYr ago with the nearby metal-poor dwarf galaxy, NGC 5253, triggering starbursts in both galaxies (van den Bergh 1980; Calzetti et al. 1999). The starburst activity in M83 is concentrated in the spiral arms and near the nucleus of the galaxy, which is heavily shrouded in dust (Gallais et al. 1991; Lundgren et al. 2008). These are the same regions of the galaxy where the high surface brightness and complex structure render the detection of novae difficult. Thus, despite our careful nova searches, the possibility that we may be missing some novae in these regions of the galaxy cannot be definitively ruled out. If such a putative population of heavily extincted novae exists, it would not be properly accounted for by our artificial nova tests, the completeness would be overestimated, and our simulations would then underestimate the true nova rate.

As discussed earlier, it was surprising to find that the observed spatial distribution of novae in M83 did not follow the background light of the galaxy. To put this result into context, it is instructive to compare the observed spatial distribution of novae in M83 with similar galaxies for which nova populations have been studied. Such galaxies include the massive, late-type, and nearly face-on spirals M51 and M101 (morphological types SA(s)bc and SAB(rs)cd, respectively), the relatively low-mass, late-type systems NGC 2403 and M33 (SAB(s)cd and SA(s)cd, respectively), and M81, a relatively early-type SA(s)ab spiral with a prominent bulge component. Unfortunately, as we explore below, a review of these studies does not suggest a simple association between the morphological type of the galaxy and the degree to which the nova distribution follows the light of the host galaxy.

In the case of the face-on, grand-design spiral M101 (Shafter et al. 2000; Coelho et al. 2008), the nova density appears to track the background light remarkably well (K-S, p = 0.94), while for M33 (Williams & Shafter 2004; Shafter et al. 2012), the cumulative distributions of the novae and the background light were only marginally consistent (K-S, p = 0.4). Similar to our result for M83, the spatial distributions of novae in both M51 (Shafter et al. 2000) and NGC 2403 (Franck et al. 2012), were also found to be more spatially extended than the background light (with 97% and 75% confidence, respectively). In theses cases, however, observational incompleteness in the central region of M51 and the spiral arms of NGC 2403 could possibly explain the discrepancy.

The situation with regard to M81 is more complex. The spatial distribution of novae in this galaxy has been studied extensively by a number of groups, all of whom have come to somewhat different conclusions about how the nova distribution compares with the galaxy light. Based on a total of 15 and 12 detected novae, respectively, Moses & Shafter (1993) and Neill & Shara (2004) argued that the M81 novae most closely follow the galaxy's bulge light, as was found in several studies of the nova population in M31 (Ciardullo et al. 1987; Shafter et al. 2000; Darnley et al. 2006). More recently, however, Hornoch et al. (2008) analyzed a much larger sample of novae (49) than either of the previous studies, finding that the nova density distribution best matches the overall (bulge + disk) background galaxy light. Given that M81 has a prominent bulge component that dominates the overall light in regions where most novae are found, the apparent discrepancy between these studies may be a distinction without much of a difference. In reviewing the earlier studies in more detail, it appears that neither the Moses & Shafter nor the Neill & Shara data are inconsistent with the hypothesis that the novae follow the overall light.

The only M81 nova study that reached a very different conclusion is that of Shara et al. (1999), who analyzed a sample of 23 novae discovered nearly a half century earlier on Palomar photographic plates taken between 1950 and 1955. They found a significant (disk) population of novae in the outer regions of the galaxy that resulted in a poor fit of the full nova distribution to the overall galaxy light (and clearly a worse fit to the bulge light). The poor fit to the overall light persisted even after the authors attempted to correct their data for missing novae in the inner bulge region of the galaxy. It is not obvious how to reconcile these early photographic data with subsequent CCD studies other than to posit that perhaps an even larger number of bulge novae than expected were missed due to the difficulty in detecting them on photographic plates when projected against the bright background of the galaxy's bulge.

Taken together, the best evidence currently available suggests that the nova distribution in M81 likely follows the overall background light of the galaxy, with novae belonging to both the bulge and disk populations. In the case of most other galaxies, the situation remains less clear. A combination of small number statistics, spatial variations in extinction (especially in late-type spiral galaxies), and limited spatial coverage (e.g., in M31) have all conspired to make it difficult to differentiate a real spatial (stellar population) variation from biases caused by observational incompleteness. As is often the case, more data will be required before we have a full and complete understanding of how the nova properties, including their specific rates, vary with the underlying stellar population.