UNCOVERING THE PUTATIVE B-STAR BINARY COMPANION OF THE SN 1993J PROGENITOR

While the COS low-resolution mode offers superior spectral resolution and lower background rates compared to similar STIS modes, its large slitless aperture inevitably integrates light from sources close to the SN; see Figure 1(d). Figure 4 plots the cross-dispersion (spatial) profile of all sources within the COS aperture, created by summing the two-dimensional images in the dispersion direction. The NUV and FUV spectra do not fall in a flat line across the detector. For this reason, a cross-dispersion profile created from all pixels in the dispersion direction will produce broad profiles which do not represent the true width of the spectrum on the detector. To balance the spatial resolution with the signal-to-noise ratio (S/N), we sum over pixels 5000–7000 in the COS/FUV/G140L/1105 spectrogram and pixels 440–520 in the COS/NUV TA image. This figure highlights that the default extraction boxes used by the HST pipeline (52 pixels for the NUV and 47 pixel for the FUV) integrate point-spread functions (PSFs) from Stars A–I in the resulting spectrum. We therefore utilize the NUV TA image to optimize our extraction parameters.

The COS NUV detector has the highest resolution of any instrument on HST, with 24 mas pixel⁻¹ and a 3-pixel resolution element in both the dispersion and cross-dispersion directions. For the NUV spectra, a very narrow (10 pixel) extraction box can isolate the SN from the other stars, but this would result in significant flux loss from the wings. Since the NUV counts are low to begin with ((1.5–3.5) × 10⁻⁴ counts s⁻¹), the additional flux from these wings proves important. Furthermore, the FUV/G140L grating has a lower resolution of 80.3 mas pixel⁻¹ and a 6 × 10 pixel resolution element (dispersion by cross-dispersion). Given the FUV resolution-element size and the S/N of the NUV data, we define the extraction box for both detectors as ∼1 FUV cross-dispersion resolution element of 11 FUV pixels (39 pixels in the NUV; see Figure 4 and Table 5). Our resulting spectrum should therefore be considered a blend of the SN plus Stars E–I, referred hereafter as "SN+Stars E–I." In Section 5 we will attempt to disentangle the different components by combining the photometry and spectroscopy with stellar models.

The dark rate in the NUV detector does not vary with detector location. To minimize contamination by other sources, we define the background region above Star C and below Star D (see Figure 4). Although the lower background region includes Star B, placing the window any lower in the NUV results in the overlapping of the background region for Stripe A with the spectrum for Stripe B. The FUV detector dark rate does vary with detector location, but these same background windows receive the minimum contamination from other sources. For consistency, we therefore define the same background regions in the FUV. It should be noted that owing to the faint SN flux, the background count rate dominates the total count rate. A visual inspection of the background at different extraction box heights confirms that the chosen background extraction box yields accurate background rates. Table 5 and Figure 4 highlight the spectrum extraction box center and height, as well as the two background region extraction box centers and heights for each central wavelength.

All extractions are performed using the COS calibration software CALCOS 2.20.1. When not using the default extraction parameters, however, two additional details must be considered. First, the flux calibration in CALCOS does not correct for the light which falls outside a nonstandard extraction box. Second, the default centering of the extraction box assumes that the target is centered in the PSA. The location of the PSA on the detector is identified with the wavelength calibration line lamp (wavecal) spectrum and a known offset between the wavecal aperture and the center of the PSA. Since the default extraction box is typically large, the centering is only required to be accurate to within a few pixels. The SN 1993J extraction requires a much smaller extraction box height, so the location of the extraction box center must be more precise.

The SN 1993J COS observations are very faint and blended, so they cannot be used to directly calculate either the aperture correction or extraction box center. Instead, we utilize the spectrum of bright, well-characterized white dwarfs used to monitor the sensitivity of both the COS FUV and NUV detectors over time. The bright and well-defined fluxes allow for an accurate measurement of the spectrum center and flux corrections for small extraction boxes. Furthermore, the large number of regular observations throughout COS operations allow us to measure the repeatability of these values.

The CALCOS x1d task was run on 14 G140L/1105 observations of WD0947+857 using the extraction parameters detailed in Table 5 for SN 1993J. Figure 5 plots the ratio of the FUV flux-calibrated spectrum with these custom extraction parameters to the CALSPEC model spectrum¹³ for each observation. A cubic spline is fit to the mean value of the observations as a function of wavelength. The uncertainty of the fit at each wavelength is calculated by fitting a cubic spline to the standard deviation of the observations around the fit. The fit to the flux ratio is used to empirically correct the flux of the extracted spectrum of SN 1993J. A similar procedure is used for each stripe of the NUV detector.

Zoom In Zoom Out Reset image size

By default, the x1d task extracts the target spectrum from the nominal spectrum location combined with the location of the aperture on the detector. However, it is possible to ask CALCOS to find the spectrum location. This option is used on the white-dwarf spectra to determine if a systematic offset exists between the nominal spectrum location and the actual spectrum location. Offsets of 2 pixels and −3 pixels are found for the G140L/1105 and G230L/2635 modes, respectively.

Each observation of SN 1993J is extracted using the parameters listed in Table 5 and the offset location found above. The different FP-POS positions of each central wavelength are combined using the fpavg task in CALCOS to create one spectrum for each central wavelength. The flux for each central wavelength is then corrected to an infinite aperture and the errors are scaled appropriately. Figure 6 plots the flux and observational uncertainty for each central wavelength. Spectra from each central wavelength are combined into a single spectrum using the splice task in the PyRAF STSDAS package. Prior to combining the central wavelengths, regions of low sensitivity and geocoronal airglow are flagged as poor quality to avoid contaminating the final spectrum (see Table 6 for a list of flagged regions).

Zoom In Zoom Out Reset image size

Figure 6 shows, however, that the estimates of the observational uncertainties associated with the combined spectrum are not well-behaved at wavelengths where the FUV and NUV spectra overlap. A detailed analysis in Section 5.3, below, further shows that over large regions of the spectrum, observational uncertainties are overestimated by varying factors depending on which central-wavelength grating is used. Instead, we find that better-behaved observational uncertainties can be obtained by not coadding spectra at wavelengths where the FUV and NUV spectra overlap. We choose to extract only the minimum observational uncertainty and associated flux at these wavelengths. The excluded regions are listed in Table 6. Figure 6 also plots the resulting uncertainties for this technique. In Section 5.3, we further show that these results are substantially more uniform over the entire spectral range, and the overall distribution is very close to normally distributed.

Figure 7 plots the final COS spectrum, which should be considered a sum of flux from the SN+Stars E–I. These are the same sources that must have been included in the Keck/LRIS spectrum presented by Maund et al. (2004). Unlabeled fainter stars are apparent in the WFC3/F336W image in Figure 1, but are not detected in the COS NUV image. We thus conclude that these stars do not contribute significantly to the extracted COS spectrum.

Zoom In Zoom Out Reset image size

Figure 8 shows a compilation of the most prominent UV lines in the COS spectrum. We also include several optical lines from the 2013 February 17 Keck spectrum (see Section 3). In this latter spectrum the Hα and [O iii] λλ4959, 5007 emission lines are strongest. Even the blended [O iii] doublet shows considerable structure, which reflects real fluctuations in the emissivity as a function of the velocity. The reality of these fluctuations is demonstrated by the same two peaks at −3472 km s⁻¹ and −4272 km s⁻¹ in both the λ4959 and λ5007 components of the [O iii] doublet. The [O iii] line shows an additional asymmetry favoring the blue side, which agrees with observations of SN 1993J by Milisavljevic et al. (2012).

Zoom In Zoom Out Reset image size

Several other lines show structures similar to that of [O iii]. In particular, the peak at ∼ − 3800 km s⁻¹ is clearly present in the Hα, Lyα, [O i], C iii], and C iv lines. Note the interstellar absorption of the Si ii λ1526 line, which distorts the C iv peak. The N ii], N iii], and Mg ii lines have a lower S/N, and for these lines the ∼ − 3800 km s⁻¹ peak is the most significant extension above the noise. Dashed lines in Figure 8 highlight the strongest peaks at both ∼ − 3800 km s⁻¹ and ∼ + 3700 km s⁻¹. These velocities also correspond to the flat part of the box-like line profiles previously identified by Fransson et al. (2005).

The asymmetry is less pronounced in the Hα line compared to the [O iii] line, although the fluctuations are present. These lines likely arise from different components, with the Hα line coming mainly from the dense, cool shell behind the reverse shock, while the [O iii] and other high-ionization lines come from the ionized outer parts of the unshocked ejecta (e.g., Chevalier & Fransson 2003). The full extension of the lines, marking the maximum ejecta velocity, is somewhat difficult to determine because of the gradual transition to the continuum. The red wing of the Hα line extends to ∼ + 9000 km s⁻¹, although the maximum velocity may be influenced by the [N ii] λ6583 line. The blue wing is blended with the [O i] λλ6300, 6364 lines, which allows us to constrain an ejecta velocity of at least ∼6500 km s⁻¹ on this side. Figure 8 also highlights these velocities with dashed lines. Compared to high-ionization lines (i.e., the C iv and [O iii]), Hα has a larger width on both the red and blue sides, with a maximum velocity of 5000–6000 km s⁻¹. As discussed above, these velocities may arise from different regions where these lines are formed. In particular, the high-ionization lines are likely to be formed by photoionization in the processed gas in the unshocked core, and are consequently expected to have lower velocity that the Hα line.

Since the SN and putative companion are spatially coincident, we must first define a template representative of the SN contribution to the overall spectrum. Prior to 2012, the most recent UV observations of SN 1993J were taken using the STIS in 2000 April with the G140L and G230L gratings (GO-8243; PI: R. Kirshner; Fransson et al. 2005). This spectrum serves as a useful reference template; we can assume that by this epoch the SN was quite evolved, and we expect little change in the spectrum besides continued fading. The validity of this assumption is highlighted by the slow UV spectral evolution in 1995–2000 shown by Fransson et al. (2005, their Figure 1) and the slow optical spectral evolution in 2000–2009 shown by Milisavljevic et al. (2012, their Figures 10 and 11). While some line strengths do vary, the underlying continuum shape and line profiles of the strongest emission lines do not change significantly, which is most important for this study.

For the 2000 STIS observations, the SN flux dominated over all neighboring stars, so contamination is not an issue. We therefore use the default pipeline product, just as Fransson et al. (2005) did. These observations are again combined using the splice task. Like the COS spectrum, wavelengths of low sensitivity and geocoronal airglow are flagged prior to the combination. A list of excluded wavelength ranges can be found in Table 7.

Figure 9 compares the 2000 STIS and 2012 COS spectra. Although the S/N is considerably lower in the 2012 spectrum, most of the lines identified in 2000 are also observed in 2012. As in 2000, the strongest line is Lyα, but it is severely affected by the strong geocoronal emission line, the geocoronal subtraction, and the damping wings of the interstellar Lyα absorption. We therefore do not include this line in our analysis, although we do note that it appears asymmetric with a strong blue wing. Both the C iv λλ1548, 1551 and C iii] λ1909 lines are still strong in the 2012 spectrum. The N ii] λλ2140,2144 doublet is weaker, but still above the noise. The N iii] λλ1747–1754 multiplet is in a noisy region of the spectrum, but the blue peak is above the noise. In Fransson et al. (2005), the feature at 2335 Å was identified with C ii] λλ2323, 2328, O iii] λλ2321, 2331, and Si ii] λ2335, and it is also well defined in the 2012 spectrum. The peaks at ∼2460 Å and ∼2640 Å are probably produced by Fe ii resonance lines. The most dramatic change is the strength of the Mg ii λλ2796, 2803 doublet, which was the strongest UV line in 2000 but is only barely above the noise in 2012.

Zoom In Zoom Out Reset image size

Figure 7 plots the combined COS NUV and FUV spectrum. We again stress that the spectrum should be considered a sum of flux from the SN+Stars E–I (see Section 4). Since the optical wavelengths are dominated by the fading SN itself, we aim here to fit the combined NUV and FUV spectrum (i.e., λ < 3000 Å). This wavelength range is not sensitive to the same stellar absorption lines observed by Maund et al. (2004) at >3600 Å (see spectral models below). To model the data, we consider all possible components with the intent of minimizing χ² (see Section 5.3).

Fading SN. As described in Section 5.1, we adopt the UV spectrum obtained with HST/STIS in 2000 (Fransson et al. 2005) as the template of the fading SN 1993J. We use the IDL MPFIT nonlinear least-squares curve fitting function (Markwardt 2009) to find the best fit by varying a single multiplicative scale factor across the entire spectrum. Figure 10 shows that a best fit can be achieved with a scale factor of 0.31 times the original STIS flux in 2000. At longer wavelengths (the NUV), the continuum levels match well. The lines, however, vary in their strength and do not appear to be well represented by a single scale factor. At shorter wavelengths, an FUV continuum excess exists which can be best fit by additional components described below.

Zoom In Zoom Out Reset image size

Stars E–H. Figure 11 plots the WFC3 photometry for SN 1993J and Stars E–H (Star I was too faint to be detected by Dolphot). The COS spectrum blends Stars E–I together with the SN, but the WFC3 photometry isolates the flux from the individual stars. For this reason, the SN 1993J photometry does not match the spectrum. The combined flux from all the stars, however, is consistent with the spectrum to within the uncertainties.

Zoom In Zoom Out Reset image size

We fit Stars E–H using Castelli and Kurucz models (Castelli & Kurucz 2004) convolved with the WFC3 throughput for each filter using IRAF's synphot.calcspec package. For simplicity, we assume the same extinction as the SN (see below). The model chosen for each star is the model for which χ² is minimized. These stellar models provide reliable estimates of the FUV fluxes despite a lack of photometry at these wavelengths. Figure 11 shows the resulting best fits. We then rescale the STIS spectrum to account for the remaining flux. Combined with the flux from Stars E–H, an STIS scale factor of ∼0.28 results in a best fit, but there is still an FUV excess implying the existence of another component.

Additional B2 star. The contribution of FUV flux from Stars E–H is minimal, and even with the addition of these stars, an FUV excess is still apparent. We therefore consider the presence of a binary companion, previously suggested to be a B2 Ia star (Maund et al. 2004; Maund & Smartt 2009). We generate a B2 Ia spectrum using the Castelli and Kurucz model in IRAF's synphot.calcspec package. We use a 20,000 K star with a surface gravity log(g/cm s⁻²) = 3.5 and metallicity [M/H] = 0.0. Following Maund et al. (2004), we renormalize to the observed V-band magnitude (V_Vegamag = 22.0 accounting for an extinction of A_V = 0.52 mag to SN 1993J). Again, we use the IDL MPFIT function to find the best fit by varying only the multiplicative scale factor for the STIS spectrum.

Figure 12 shows an improved fit with an STIS scale factor of ∼0.12. This model, however, is not ideal. The combined flux from the scaled STIS spectrum and B-star model (orange) should be consistent with the observed WFC3 photometry, but in this case the B-star model is too bright. To optimize our fit, we consider other stellar types and luminosity classes ranging from 10,000 K to 31,000 K, as well as normalization magnitudes (although we do not vary the metallicity or surface gravity values). We minimize χ² (e.g., maximize the probability function) with a temperature of 24,000 K renormalized to V_Vegamag = 22.9 mag (see Figure 13). Given the noise associated with the spectrum, the data are somewhat degenerate with several models that include B stars in the range 19,000–27,000 K (B3–B1). We thus infer the presence of a spectral component consistent with continuum from a hot B star.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To effectively utilize the reduced χ² requires a detailed understanding of the observational uncertainties. CALCOS, the COS calibration pipeline, generates both a flux-calibrated one-dimensional spectrum and an associated uncertainty. This uncertainty is calculated using the formula given by Gehrels (1986; see COS Data Handbook; Massa & York 2013). Gehrels' formula, however, offers only an upper limit to the uncertainty regardless of the number of source counts (for large counts the formula approaches the typical $\sqrt{n}$ error approximation).

Figure 6 shows two methods used for generating our spectrum of SN 1993J. In the first, we combine all spectra from each central wavelength into a single spectrum. In the second, we do not coadd spectra at wavelengths where the FUV and NUV spectra overlap. We choose to extract only the minimum observational uncertainty and associated flux at these wavelengths. For each case, the residual errors can be characterized by the distribution of χ = (Observed(i) − Model(i))/Uncertainty(i), for which a normal distribution should be characterized by a Gaussian with a standard deviation equal to 1. For the two techniques described here, Figures 14(a) and (b) show the values χ as a function of wavelength and the resulting histogram.

Zoom In Zoom Out Reset image size

Figure 14(a) varies significantly, while Figure 14(b) shows a more normal distribution. Still, the uncertainties appear too large in the FUV (i.e., χ < 1). To compensate, we decrease the size of all uncertainties in spectral elements with a wavelength shorter than 1660 Å by a factor of 2.9, and all spectral elements with a wavelength longer than 1660 Å by a factor of 1.4. Figure 14(c) shows the results, which is nearly a normal distribution. We adopt these uncertainties throughout the modeling in this paper.

Despite the improved accuracy of the uncertainties, the resulting reduced χ² values cannot be considered meaningful on an absolute scale because the degrees of freedom are not well defined. We therefore calculate the probability (Q) that the resulting χ² value is due to chance and normalize relative to the sum of the resulting probability distribution (see Figure 15). We consider models summing to 68% of the combined probability (e.g., 19,000–27,000 K) to be the likely range of models that can fit the COS spectrum.

Zoom In Zoom Out Reset image size

While we do find, like Maund et al. (2004), that the surrounding stars contribute little to the overall flux, we also consider the possibility that this hot B-star emission may not arise from the supposed binary companion. First, we cannot definitively rule out line-of-sight coincidences. Moreover, the high-resolution TA image contours in Figure 4 reveal the presence of an unresolved UV source just south of the SN position. We also see a hint of this feature in the WFC/F336W filter in Figure 1. At a distance of 3.6 Mpc, this angular separation of >01 corresponds to >1.5 pc, thereby ruling out the possibility of this source being the putative companion of the progenitor. Furthermore, the flux of this unresolved source is low relative to the SN+putative companion. The presence of this emission, however, suggests the possibility that the local progenitor environment is crowded. If this is true, the B2-star detection within the HST PSF does not necessarily have to be the binary companion. Future HST imaging may be able to probe the progenitor environment in more detail once the SN sufficiently fades.