Discovery of 20 UV Emitting SNRs in M31 with UVIT

The list of SNRs and SNR candidates within M31 was obtained from existing optical, X-ray and radio SNR lists¹¹1Galvin & Filipovic (2014) gives a catalog of 916 radio sources in M31 detected at 20cm, however the sources which are SNRs are not identified.. The optical SNRs were from Lee & Lee (2014), which contained 156 SNR candidates observed using H$\alpha$ and S[II] images. The X-ray SNRs were from Sasaki et al. (2012) which lists 26 confirmed SNRs and 21 SNR candidates. These sources were first catalogued in an XMM-Newton survey of M31, which catalogued a total of 1897 X-ray sources (Stiele et al., 2011). The radio SNRs were from Braun & Walterbos (1993) which lists 24 high confidence ($>5\sigma$) and 6 medium confidence (3 to 5$\sigma$) detections at 1465 MHz. That work found 52 SNRs and candidates using narrow band imaging in [SII] and H$\alpha$ filters of the NE half of M31, then matched those with radio continuum imaging of M31.

We combine the three sets of SNRs. The 52 optically-detected SNRs (including 30 with radio emission) from Braun & Walterbos (1993) were included in the study of Lee & Lee (2014). Four of the 30 with radio emission were re-examined by Lee & Lee (2014) and found not to be SNRs, and one more was found by us to be contaminated by stellar emission. This left 25 SNRs from Braun & Walterbos (1993) that are detected in radio, which are included in the list in Lee & Lee (2014). Some SNRs are listed in both optical and X-ray catalogues as noted in the analysis of X-ray SNRs in the northern disc of the M31 (Sasaki et al., 2018). Using an angular separation of $18^{\prime\prime}$, we found an additional 18 source matches between the radio and X-ray SNR catalogs. This left 179 unique SNRs or SNR candidates, hereafter referred to as SNRs, within M31. However, two of these are outside the M31 UVIT survey region, leaving 177 unique sources for analysis using UVIT data: 119 detected only in optical, 22 detected only in X-ray, 13 detected in optical and radio, 11 detected in optical and X-ray, and 12 detected in optical, X-ray and radio. The location of these 177 SNRs in M31 for the different categories is shown in Figure 1(b).

The locations of the 177 SNRs were used to determine within which field each object was located and thus which filters were oberved for each SNR. Next, we carried out a set of tasks to find those SNRs which have diffuse emission not too contaminated by UV emitting stars in M31. SNRs in UV are characterized by diffuse emission, in contrast to stars which are unresolved point sources. UVIT has high enough spatial resolution in most cases to separate diffuse from the stellar point-source emission.

An initial inspection of the 177 SNRs was undertaken to remove any sources from the list without any UV emission within the optical radius of the SNR, given in Lee & Lee (2014). For the 22 X-ray-only SNRs, no optical SNR radius was available, and neither was an X-ray radius, so the position error was used instead. The radius of analysis around each source (whether from optical or X-ray) is henceforth referred to as the “SNR radius”. Several SNRs were contained within densely packed, UV-emitting star clusters, so that no SNR emission could be distinguished from the stellar emission. The “no emission” and crowded sources were removed from the list of SNRs for analysis, leaving 126.

Further inspection of these 126 SNRs often revealed stars within the radii of the SNRs. Thus, we search for known stars within the SNR radius of each SNR: Vizier (https://vizier.cds.unistra.fr/) was used to search through stellar catalogues. A number of stellar catalogues were initially searched, but most were contained within the GAIA Early Data Release 3 (EDR3) catalogue (Bailer-Jones et al., 2021). We found all EDR3 stars within a 15″search radius (15″was the largest optical radius of any source). A total of 766 stars were found in the EDR3 data (many would not have UV emission), and their coordinates were imported into CCDLAB to search for stellar contamination. 42 sources were removed from our SNR list because all UV emission within the SNR radius was identified with catalogued GAIA stars, leaving 84 sources in our list of SNRs. This process was repeated using the full data release (DR3) upon its publication on June 13th, 2022. 5 additional stars were determined to be nearby the SNRs, but none associated with the remaining 84 with UV emission.

The 84 sources were further analyzed by searching for stellar sources from additional stellar catalogues. A number of catalogues were searched during this process, including a BVRI analysis of M31 objects using the McGraw Hill Telescope, a Swift/UVOT source catalogue, and a XMM-OM object survey (Magnier et al., 1992; Yershov, 2014; Page et al., 2012). Vizier searches from these catalogues did not provide additional stellar contaminants in the 84 SNRs. However, the catalogue of M31 supergiants from the local group galaxy survey (Massey et al., 2016) provided additional stellar sources within a 15″search radius of the 84 SNRs. The red supergiant catalogue of Massey et al. (2021) was also included in the Vizier search. 610 supergiants from these two catalogues were found to be nearby our remaining SNRs: most of these were blue supergiants or luminous blue variables. This process led to a number of UV emissions to be reclassified as supergiant stars. 43 sources were removed from our SNR list, leaving 41 SNRs. One more source was removed using the HST M31 PHAT catalogue (Williams et al., 2014) using PHAT stars with F275W magnitude brighter than 20.

Out of the 40 SNRs that remained, 7 had UV emission with no associations with stars. The other 33 consisted of either i) sources with diffuse emissions that overlapped with GAIA stars or Massey supergiants or ii) diffuse emission too dim to be reliably measured. Stars within diffuse emission were examined to determine if they were likely to emit UV radiation: Larger U-B values indicated that the star emitted very little UV radiation, and would not contaminated the UV measurements. These 33 SNRs were analyzed in CCDLAB to determine whether or a measurement of the diffuse emission could be done. 13 of these were isolated enough from nearby UV sources to analyze using box measurements. 5 sources had clear indications of diffuse emission, but the region had too many overlapping stars (i.e. was confused) for a reliable flux measurement to be taken. The remaining 15 sources were removed from the list as there was either no clear indication of SNR emissions within the SNR radius, or the measured flux was too dim compared to the background to result in a reliable measurement.

The result of the above selection process left 25 UV emitting SNRs consisting of 7 without stellar contamination; 13 with stellar contamination which can be separated from the diffuse emission and 5 more with likely diffuse emission but too confused with stars for measurement. The locations of the 20 SNRs in M31 with clear diffuse UV emission and 5 SNRs with likely diffuse emission are shown in Figure 2. We do not consider further the 5 diffuse but confused sources for which reliable UV fluxes could not be obtained. Thus, the number of SNRs for which we can carry out flux measurements is 20.

Properties of the 20 SNRs with detected diffuse UV emission are shown in Table 1. Table 1 lists each SNR’s ID numbers, including their original Lee and Stiele (SPH11) IDs, their J2000 coordinates, the optical diameter and error, the likely progenitor type, and their optical and/or X-ray luminosities.

From the measured fluxes and the known distance to M31, we determine the SNR UV luminosities in the different filters. CCDLAB provides source fitting methods for photometric measurements (Postma & Leahy, 2017). The methods include Gaussian and Moffat functional fits, a Curve-of-Growth fit (COG), and a box fit. For diffuse emission Gaussian and Moffat functions, designed for fitting point sources, are not suitable. The COG fit works well to measure all counts within a given radius if there is a good background outside that radius, but does not work well for our diffuse sources usually surrounded by stellar sources. Thus, we use the box method to calculate the fluxes of diffuse emission and to subtract background emission while avoiding stellar emission in both source and background boxes.

The box method yields the total photon counts in a chosen box, which can be square or rectangular. The SNR source counts were measured using different box sizes, generally 4 different sizes from 17 by 17 pixels to 23 by 23 pixels. However, more crowded sources would be measured 2 or 3 times using smaller boxes, and some sources required larger boxes due to their unusual shape. A few sources were large and irregularly shaped. These required a larger box to encapsulate the UV emission, and for these cases small boxes would be used to remove the counts from any stars that fell within the larger box. In order to account for background variations, 6 different measurements were done for each filter using manually chosen boxes. The average of the 6 measurements was used as the background, and the standard deviation of the 6 measurements used as the uncertainty.

The counts for each filter band image were divided by the net exposure time of each image to obtain a count rate. This was corrected for the extended wings of the point-spread-function as given in Table 5 of Tandon et al. (2017). Then the count rate was converted to flux in units of erg s^-1 cm^-2 $\AA^{-1}$ using the flux conversions given in Table 4 of Tandon et al. (2017). Additional uncertainties include Poisson photon counting errors, uncalibrated spatial variations in detector sensitivity, and uncertainties in correction for the wings of the point spread function. These were included in quadrature in the calculation of flux errors. The fluxes and uncertainties were converted into luminosities using the distance to M31 and the effective bandwidth of each filter. The filter effective wavelengths and effective bandwidths are given in Table 3 of Tandon et al. (2017).

The images of the 20 M31 SNRs in the F148W band are shown in Figure 3. These sources exhibit diffuse emission which is not associated with stars, although the strength of the diffuse emission varies. Images of two of these sources (ID 3 and ID 6) in all five UVIT filter bands are shown in Figure 4. The catalog of 20 SNRs in M31 with luminosities in the detected UVIT bands is given in Table 2. Because of some overlap of the fields, some SNRs were imaged in two different fields in the same filter. The reported band luminosity and uncertainty for those are the average of these measurements.

The spectral shape of each source is shown in Figure 5, panels (a) and (b). The F148W filter luminosity is the largest for all SNRs analyzed, in part caused by the wider effective bandwidth for the F148W filter. The other filter band luminosities vary in prominence, however the spectral shapes of the 20 M31 SNRs are similar.

The spectra of known UV-emitting SNRs shows that their emission in the UVIT filter bands would be dominated by emission lines, with a small contribution from continuum radiation. A list of UV emission lines measured in known SNRs (Fesen & Hurford, 1996) is presented in Table 3, together with the UVIT filter bands that contain these lines. These SNRs include 3 Galactic SNRs- the Cygnus Loop (Raymond et al., 1980, 1981, 1988), Vela (Raymond et al., 1981), and Puppis A (Blair et al., 1995); 3 SNRs in the Large Magellanic Cloud- N132D (Blair et al., 2000), N49 (Vancura et al., 1992), and N103B (Blair et al., 2020); and 1 SNR in the Small Magellanic Cloud- E0102-7219 (Blair et al., 1989, 2000). Table 4 lists these SNRs, with SNR Type, distance, and aperture size for the line flux measurements given in the references.

We estimated the UVIT band luminosities for the known SNRs by summing the published luminosities or fluxes (converted to luminosities) of the detected emission lines which fall within each UVIT filter, as listed in Table 3. The results of this are given in columns 5 through 9 of Table 4 and the band luminosities vs. the effective wavelengths of the UVIT filter bands are shown in panel (a) of Figure 6. Although the band luminosities of the different sources are quite variable, the F148W luminosity is the brightest. That band includes more emission lines than the other bands (see Table 3).

Supernovae and SNRs can be categorized into two separate types: thermonuclear runaway (Type Ia) and core collapse (CC). The small sample of known SNRs with UV spectroscopy includes only one type Ia SNR (N103B). This SNR has band luminosities (panel (a) of Figure 6) which does not appear different to the CC-type SNRs. Scaled band luminosities are defined as the band luminosities for each SNR divided by the F148W band luminosity for that SNR. The scale band luminosities of the seven known UV-emitting SNRs are shown in panel (b) of Figure 6. They are all quite similar, with factor $\sim$2 variations between the different SNRs.

Although UV spectroscopic data for the M31 SNRs does not exist, we can compare the UV emission of the M31 sources to known UV-emitting SNRs by comparing the UVIT band luminosities. For several of the known SNRs, the aperture only covered part of the area of the SNR, so that we do not have a good measurement of the line luminosities over the whole SNR. The extrapolation from the aperature to the whole SNR is a highly uncertain factor because of line brightness variations over the face of the SNR. However, if the line ratios are nearly constant over the face of the SNR, the scale band luminosities should be representative of the spectral shapes of the whole SNR. Thus we compare the spectral shapes of the known SNRs to one another and to the M31 SNRs using the scaled band luminosities.

For the band luminosities, known SNRs (panel (a) of Figure 6) show larger variations than the 20 new UV emitting SNRs in M31 (panels (a) and (b) of Figure 5). This is, in part, caused by the different fractions of the SNR area measured in the aperture of the spectroscopic observations (Table 3). However, the scaled band luminosities (panels (b) of Figure 6) show that the spectral shapes of the known SNRs are quite similar. The F148W value is the largest, followed by the F169M value, then with similar values in the remaining 3 bands (F172M, N219M and N279N). The scaled band luminosities of the 20 M31 SNRs (panels (c) and (d) of Figure 6) have remarkably similar variations to the known SNRs. The strong similarity of the rescaled band luminosities of the newly detected 20 M31 SNRs to known UV-emitting SNRs (Figure 6) is evidence that the detected UV emission is from the SNR, rather than from foreground or background objects.

This current list of 20 UV-emitting SNRs in M31 is likely incomplete. The interstellar extinction of UV light from M31 by the Milky Way’s ISM is small, but the ISM of M31 can have significant extinction. The extinction has been measured for stellar clusters in M31 using UVIT photometry of stellar clusters (Leahy et al., 2022b): the mean extinction is $E(B-V)$=0.24 and ranges from 0 to 0.6. Thus SNRs within or on the farside of M31’s disc may be undetected because of extinction. More than half (Section 3.2) of the SNRs from the original list of 177 were excluded because of confusion with stellar emission: there was not enough diffuse emission separated from overlapping stars within the SNR radius for detection.

To determine density of environment and evolution status of SNRs, X-ray observations of the thermally emitting shocked gas are required. There are 6 SNRs in M31 which have had their X-ray spectra analyzed with hot plasma models from shocked gas, from Sasaki et al. (2012) and Sasaki et al. (2018). We use the spectral parameters for SPH11 SNRs 1050 and 1066 from Sasaki et al. (2012) and for SPH11 SNRs 1234, 1275, 1535 and 1599 from Sasaki et al. (2018)²²2SPH11 1234 was analyzed in both: we used the newer analysis from the later reference.. The emission measures ($EM$) for each SNR was calculated from ”norm” or from the flux, if no ”norm”, using XSPEC and the best fit spectral model.

The modelling software used is SNRpy (Leahy et al., 2019; Leahy & Williams, 2017) which is based on the unified models of SNR evolution of Truelove & McKee (1999), with extensions added, including non-equilibrium ionization. For each SNR, we fit the measured shocked-gas temperature $kT$ and emission measure $EM$ for 3 different cases: i) explosion in a uniform environment (s=0) and emission from the forward-shocked gas; ii) explosion in a stellar wind (s=2) and emission from the forward-shocked gas; and iii) explosion in a stellar wind (s=2) and emission from the reverse-shocked gas. Explosions in a uniform medium lead to much brighter emission from forward-shocked gas, but explosions in a wind environment can lead to either brighter emission from the forward-shocked gas or from the reverse-shocked gas. The results of the models are given in Table 5. Because the measured X-ray spectrum is for the brighter component (i.e. forward or reverse-shocked), we mark the models which are consistent with the observations (i.e. measured $EM$ larger than $EM_{2}$) in the Table. Emission from the forward-shock gas in a uniform medium (s=0) or emission from the reverse-shocked gas in a stellar wind (s=2) are consistent with observations. Other information can help to select a preferred model for each SNR. E.g., SNRs with energy above $10^{52}$ erg should be rare (Leahy & Filipović, 2022c; Leahy et al., 2020b; Leahy, 2017). This disfavors the (s,n)=(2,12) reverse-shock models for SPH11 1234, 1535 and 1599. The energies, ages and densities (or stellar wind parameters) are reasonable for the other models.

The 2 UV-emitting SNRs with X-ray spectra, SPH11 1066 and 1275, are marked with (UV) in Table 5. The clear difference between the UV emitting SNRs and the UV non-detected SNRs is that the 2 UV-emitting SNRs are Type Ia. Because of the large age of Type Ia progenitors compared to CC progenitors, Type Ia are expected to be at significant height above the disk plane, and thus to have low extinction. The models for the 6 SNRs have a range of ages, explosion energies and densities, and the sample is too small to see systematic differences in the ages, densities or explosion energies between the UV-detected and non-detected SNRs. One would expect the column density (given in Sasaki et al. 2012 and Sasaki et al. 2018) to be smaller for the UV-detected SNRs. The column densities are small, with large uncertainties, except for SPH11 1599 which is not detected in UV. Better determination of column densities with better X-ray spectra are needed to confirm the relation between UV-detection and low column density.

The 177 SNRs in M31 were detected in different wavebands; optical, X-ray and radio, prior to the current work. Panel (a) of Figure 7 shows the numbers of SNRs detected in single and multiple wavebands using a Venn diagram. Pannuti et al. (2000) shows similar Venn diagrams (optical, radio, and x-ray detected SNRs) for the nearby galaxies M33 and NGC300. M33 has a total number of 109 SNRs and SNR candidates and NGC300 has a total of 44, compared to the total of 177 SNRs and SNR candidats for M31. For all 3 galaxies (M31, M33 and NGC300), optically-detected SNRs dominate with fractions of detected SNRs of 155/177, 79/109 and 28/44, respectively. The X-ray and radio detections are significantly smaller fractions: the X-ray detected fractions are 45/177, 21/109, and 6/44 respectively; and the radio detected fractions are 25/177, 53/109 and 17/44, respectively. The optically detected fractions are similar ($\sim 0.6-0.9$) for all 3 galaxies; the radio detected fractions higher for M33 and NGC300 ($\sim 0.4-0.5$) than for M31 ($\sim 0.15$), and the X-ray detected fraction higher for M31 ($\sim 0.25$) than for M33 and NGC300 ($\sim 0.15-0.2$). This is probably the result of the intensive X-ray observations of M31 (Sasaki et al., 2012, 2018) compared to the other two galaxies. The three sets of optically-detected, radio-detected and X-ray detected SNRs show little overlap for all three galaxies. This is consistent with the opposing selection effects for these 3 wavebands, as discussed in Pannuti et al. (2000): SNRs identified through optical represent those located in regions with relatively low confusion from H$\alpha$ emission, well away from star-forming regions; radio-selected SNRs are biased toward star-forming regions; and X-ray SNRs are selected for soft X-ray spectra and association with H II regions, so are biased against SNRs with hard spectra and no optical counterparts.

Bozzetto et al. (2017) considers the statistics of optical, radio and X-ray detections for the Large Magellanic Cloud (LMC). The Venn diagram of SNRs for the LMC shows most SNRs (47 of 59) are detected in all three bands. This is probably a result of sensitivity to luminosity because of the closer distance of the LMC ($\sim$20 times closer than the other nearby galaxies discussed above). The Venn diagrams given in Bozzetto et al. (2017) include NGC300 and M33, with similar numbers to those given in Pannuti et al. (2000), and M31 with results similar to Figure 7 here. The other galaxies with Venn diagrams in Bozzetto et al. (2017) are the Small Magellanic Cloud (SMC), NGC7793, NGC 6946, NGC 55, and one diagram for NGCs 2403, 3077, 4214, 4395, 4449 and 5204 combined (hereafter referred to as ”NGCcombined”). For the SMC, which is also nearby, the diagram is similar to that for the LMC with most SNRs detected in all 3 bands. The diagrams for NGCcombined, NGCC 7793 and NGC 6946 are dominated by SNRs detected in optical only, like M31, M33 and NGC 300. NGC 55 has too few SNRs (6 total) to draw conclusions on numbers. The differences are likely the result of the differing sensitivities of the observations in the different wavebands for each of the galaxies, as discussed by Bozzetto et al. (2017).

The 20 UV-emitting SNRs in M31 are listed in Table 1. The other wavebands in which these 20 SNRs are detected are as follows. Sources 1, 2, 4, 5, 8, 10, 11, 12, 14, 16, 17, 18, 19, and 20 were listed only in Lee & Lee (2014) (optical); sources 7 and 9 were listed in Lee & Lee (2014) and in Braun & Walterbos (1993)(optical & radio); source 15 was listed only in Sasaki et al. (2012) (X-ray) and sources 3, 6, and 13 were listed in all 3 (optical, X-ray, & radio). Panel (b) of Figure 7 shows the numbers of UV-emitting SNRs detected in optical, X-ray and radio. Nearly all of the UV-emitting SNRs (19 of 20) are detected in optical. This is not surprising because the emission mechanisms for UV and optical are most similar: forbidden and recombination lines from shock-ionized gas. In contrast, the radio emission mechanism is synchrotron from shock-accelerated electrons and the the emission mechanism is primarily thermal bremmstrahlung with some contribution from lines.

The total number of UV-emitting SNRs in M31 is similar to the number radio emitting SNRs (20 vs. 25) but much smaller than numbers of optical or X-ray emitting SNRs. The fractions of UV-emitting SNRs in M31 are: 19/155 (optical), 5/25 (radio) and 4/45 (X-ray). These have not been measured for other galaxies yet, but are similar ($\sim 0.1-0.2$) for the 3 different categories. This indicates that the UV-selection criterion is different than the optical, radio and X-ray criteria listed above. The most important expected UV detection criterion is source extinction. This is important for the 7 previously known SNRs, which all have low extinction (see references in Table 4). The UV extinction to SNRs in M31 is dominated by line-of-sight distance through M31’s disk³³3The Milky Way contribution to extinction is small in the direction of M31. For M31 SNRs detected in optical, radio and X-ray, the disk extinction is small at those wavelengths so they should be detected independent of distance into the disk. The SNRs detected in UV will be those on the near side of the disk. The fraction detected in UV to other wavebands should be determined by extinction, i.e. disk geometry, and thus approximately constant and equal to the fraction of disk which is on the near side of M31 and with low extinction in UV. The visible to UV extinction curve (Fitzpatrick & Massa, 2007) has E($\lambda$-V)/E(B-V) $\sim$3.5 to 7 (average $\sim$5) for $\lambda=$120 nm to 280 nm (with the peak at 220 nm). Typical measured E(B-V) values in M31 are in the range 0.0 to 0.6 (Leahy et al., 2022b), which yields E($\lambda$-V) from 0 to 3 mag, and extinction factors from 1 to 0.06. The fraction of UV to optical SNRs in M31 (19/155$=0.12$, Figure 7) is consistent with that expected from extinction, but the differing sensitivity of UV and optical observations could also affect the fraction.