EXPANDED VERY LARGE ARRAY OBSERVATIONS OF GALACTIC SUPERNOVA REMNANTS: WIDE-FIELD CONTINUUM AND SPECTRAL-INDEX IMAGING

In order to test the wide-band imaging capabilities of the EVLA,³ we have observed several known and possible Galactic supernova remnants (SNRs). There are 274 Galactic SNRs cataloged (Green 2009), the majority of which are classified as "shell" remnants, showing limb-brightened synchrotron radio emission, typically with S∝ν^∼−0.5. Some SNRs are classified as "filled-center" remnants (or "plerions"), as they show emission at radio wavelengths that are brightest in the center, due to the power input from a central pulsar. These filled-center remnants generally have much flatter radio spectra at GHz frequencies, with S∝ν^{∼−0.0–0.3}. Finally, there are some remnants that are classified as composite remnants, as they show a mixture of limb-brightened shell-like emission, with a filled-center-like core (i.e., a pulsar wind nebula) showing a much flatter spectrum. In addition to the known remnants, there are many other possible and probable Galactic SNRs that have been proposed, for which additional observations are needed to confirm their nature. In many regions of the Galactic plane, particularly near to b = 0° and l = 0°, radio emission from SNRs is easily confused with thermal radio emission (typically with S∝ν^∼−0.1) from H ii regions. The wide band of the EVLA allows a variety of spectral studies of SNRs to be made, in order to: (1) investigate spectral variations across the face of shell remnants to study the differences in the shock acceleration mechanism at work in different regions of the remnants, (2) study the flat spectrum cores of composite remnants, and (3) clarify the nature of proposed possible SNRs by disentangling the non-thermal synchrotron emission from confusing thermal emission.

We have observed two smaller known remnants G16.7+0.1 and G21.5−0.9. G16.7+0.1 is a "composite" remnant, ∼3' across (e.g., Helfand et al. 1989). G21.5−0.9 was classified as a filled-center remnant, only ∼15 in extent, until a faint X-ray shell ∼4' in diameter was identified by Slane et al. (2000). Since then it has been classified as a composite remnant, although a radio counterpart to the X-ray shell has not been identified (see Bietenholz et al. 2011). G21.5−0.9 contains a pulsar, which has a high spin-down luminosity (Gupta et al. 2005). We also observed a field centered at l = 196, b = −02, which contains a "high-probability" SNR candidate, G19.66 − 0.22, identified by Helfand et al. (2006). Finally, we observed one larger known but poorly studied shell remnant, G55.7+3.4, which is ∼20 × 25 arcmin² in extent (see Goss et al. 1977). An old, probably unrelated pulsar lies within the northern edge of this remnant.

The observations were made using the EVLA in D-array. The L-band observations used the full 1 GHz bandwidth covering the frequency range of 1–2 GHz. The C-band observations used 2 GHz bandwidth in two separate ∼1 GHz bands. The observation parameters for all the fields are listed in Table 1. At the time of observations, not all antennas in the array were outfitted with the new L- and C-band receivers. Due to this and other failures during the commissioning phase of the EVLA, typically 20 antennas were used for these observations. The L-band data was recorded in 8 or 16 spectral windows (SPWs) at a frequency resolution of 2 and 1 MHz, respectively, with an integration time of 1 s. The C-band data was recorded in 16 SPWs at a resolution of 2 MHz. High time and frequency resolution was required to allow effective removal of radio frequency interference (RFI), accurate bandpass calibration, and corrections for wide-band group delay present in the data from the EVLA WIDAR correlator. To reduce the data volume, after initial data flagging the data were averaged to 10 s resolution in time. The standard flux calibrators 3C286 and 3C147 were observed for flux calibration and the compact sources J1822−0938 (for G16.7+0.1, G19.6−0.2, and G21.5−0.9) and J1925+2106 (for G55.7+3.4) were observed at intervals of 30 minutes for phase and bandpass calibration. Using a known frequency-dependent model for the flux calibrators (including spectral index), antenna gains for each frequency channel across the observed bandwidth were determined for flux and group delay calibration. The antenna gains for each SPW were also determined using the phase calibrators. In order to account for the spectral index of the calibrators, a model including the frequency-dependent flux was determined using the multi-scale multi-frequency synthesis (MS-MFS) imaging algorithm which simultaneously makes Stokes-I and spectral-index maps. The final complex bandpass shape was determined using this model and applied to the data for the target fields. This calibration procedure allowed calibration of time-varying direction-independent gains without transferring the effects of the frequency-dependent flux of the calibrators to the target fields.

RFI-affected frequency channels were flagged throughout the observing band. Full SPWs covering the frequency range 1.516–1.643 GHz had to be dropped due to the presence of strong RFI in the L band. Ten channels at the edges of the SPWs were also flagged due to the roll-off of the digital filters in the signal chain. The rest of the frequency channels were used for wide-band imaging.

Conventional image reconstruction techniques for interferometric imaging fundamentally models the brightness distribution as a collection of scaleless components (amplitude per pixel) leaving errors that limit the imaging performance. Scale-sensitive methods (Bhatnagar & Cornwell 2004; Cornwell 2008) significantly reduce the magnitude of such errors by explicitly solving for the scale size of emission across the field of view. Imaging performance using wide-band interferometric observations of fields with extended emission is additionally limited by the fact that not only does the scale of emission vary across the field of view, the spectral properties also vary across the field. The dominant source of these spectral variations is the change of antenna primary beam with frequency and spatial dependence of the spectral index of the radio emission. Both effects result in variation of the strength of emission as a function of frequency in a direction-dependent manner. Furthermore, antenna primary beams are typically rotationally asymmetric which result in time-varying gains due to the rotation of the primary beams with parallactic angle for El-Az mount antennas. Scale-sensitive imaging reconstruction algorithms that can also simultaneously account for time- and frequency-dependent effects are therefore required for wide-band wide-field imaging with the EVLA, particularly in the Galactic plane. For this, two new algorithms, namely the MS-MFS (Rau 2010; Rau & Cornwell 2011) and the A-Projection (Bhatnagar et al. 2008) algorithm, have been developed. The MS-MFS algorithm combines the scale-sensitive deconvolution with explicit frequency-dependent modeling of the emission to map the spatial distribution of the spectral index in a scale-sensitive manner. The A-Projection algorithm on the other hand corrects for the time and frequency dependence of the antenna primary beams as part of the iterative image reconstruction.

We used the MS-MFS (Rau 2010; Rau & Cornwell 2011) algorithm implemented in the CASA⁴ post-processing package for wide-band imaging. The spectral-index information derived using the traditional method of using images made at different frequencies make spectral-index images at the lowest resolution and at the sensitivity of single-channel bandwidth. Spectral-index images made using the MS-MFS algorithm, on the other hand, do not require smoothing the images to the lowest resolution and produce the spectral-index images at the full sensitivity and resolution offered by the data. This must be combined with techniques to correct for various time-variable antenna primary beam (PB) effects during image deconvolution to achieve noise-limited imaging performance. While the A-Projection algorithm can be used to correct for these PB effects during image deconvolution, for the images presented here we used only the time-averaged PBs (see Figure 3) to apply correction post deconvolution. Work toward a fully integrated wide-field wide-band imaging algorithm is in progress.

4.1. G16.7+0.1, G21.5−0.9, and G19.6−0.2

Figure 1 shows the Stokes-I and spectral index images of two cataloged SNRs (G16.7+0.1 and G21.5−0.9; Green 2009) and one candidate SNR (G19.6−0.2). The top row shows the Stokes-I images and the bottom row shows the spectral-index images, with the Stokes-I contours overlaid on the spectral images. Correction for the primary beam response raises the noise away from the center. In order to show the detailed brightness distribution, we chose to show the Stokes-I images without correcting for the primary beam response. The integrated flux reported below was determined from PB-corrected images. The spectral-index images have been corrected for the average primary beam effects, since the effect of primary beam response in the spectral-index images is stronger. Image of all the fields covering the full wide-band sensitivity shows many more compact and diffused sources. Due to space limitation, here we show central portions of these images.

Zoom In Zoom Out Reset image size

The first column has the images at a reference frequency of 6 GHz of the composite SNR G16.7+0.1 at a resolution of 10''. The SNR has a flatter spectrum core with an average spectral index of −0.54 ± 0.02 (S∝ν^α) surrounded by a relatively steeper spectrum nebula with the spectral index ranging from −0.8 to 1.1. The total continuum integrated flux is 1.2 Jy with an rms noise of 11 μJy beam⁻¹. The diffused sources east of the SNR are unrelated but real sources of emission. The images in the second column are of the filled-centered SNR G21.5−0.9, also at the reference frequency of 6 GHz. The rms noise in this image is ∼30 μJy beam⁻¹ and the resolution is 10''. The spectral index across this SNR is uniform across with a value of +0.12 ± 0.03 and an integrated flux density of 6 Jy. Also visible in the image is the compact source called the "northern knot" about 2' to the north. The long extended feature extending toward the northwest direction is part of the separate cataloged SNR G21.6+0.84. The third column contains the L-band images at a reference frequency of 1.5 GHz, of the region centered on the Galactic coordinates l = 196, b = −02. This field contains a candidate SNR G19.66−0.22 (Helfand et al. 2006). Stokes-I image shown in the top row shows weak diffused emission surrounding two compact strong sources, making it difficult to determine the nature of the objects based on Stokes-I morphology alone. The spectral-index image, however, clearly separates sources of thermal and non-thermal emission. The spectral index of the diffused emission ranges from −0.4 to −0.5. The spectral index of the compact source (G19.67−0.15) toward the northern edge of the diffused nebula has a spectral index of −0.3. The superimposed compact thermal source near the center is the known ultra-compact H ii region G19.61−0.23 (e.g., Wood & Churchwell 1989; Hofner & Churchwell 1996) also with a spectral index of +0.3. Based on morphology of the spectral distribution we conclude that this field contains an SNR, possibly of filled-centered type.

4.2. G55.7+3.4

The SNR G55.7+3.4 is classified as an "incomplete-shell SNR" with a pulsar within the boundary of the remnant based on the only existing radio image at 610 and 1.415 GHz by Goss et al. (1977). These are the only images published in the literature of this SNR made over three decades ago using the Westerbork Synthesis Radio Telescope. The full 2° × 2° Stokes-I L-band image centered on this SNR at the reference frequency of 1.5 GHz covering the full wide-band sensitivity pattern of the EVLA antenna is shown in Figure 2. This is the highest resolution and sensitivity image of this region to date. The wide-band sensitivity pattern of the EVLA extended significantly beyond even the first sidelobe of the PB at the lowest frequency in the band. The time-averaged wide-band sensitivity pattern of the EVLA for the equivalent frequency range and field of view is shown in the left panel of Figure 3. The one-dimensional slice through this pattern in the right panel shows the sensitivity pattern extending significantly beyond the "main lobe" at the level of 5%–15% of the peak value. With sources, some strong, detected up to 1° away from the center of the image, imaging the full field was required to reach noise-limited imaging at the center of the field. Due to this wide field of view, even for D-array observations, correction for the effects of non-coplanar baselines had to be applied. This was done using the W-Projection algorithm (Cornwell et al. 2008). The Stokes-I image of the SNR shows a nearly complete shell-type SNR. At a resolution of 30'', this image shows for the first time a network of filamentary structures along with smoother diffused emission filling the boundary of the shell. The spectral-index image of this SNR, however, was not reliable at all scales since the large scale emission is not well constrained by the shortest measured baselines at all frequencies. The filaments, however, show up as regions of steeper spectral index compared to an average spectral index of −0.6 of the surrounding regions. The strong compact source (G55.78+3.5) on the northern edge of the shell in the Stokes-I image is an older unrelated pulsar (PSR J1921+2153). In the spectral-index map (not shown here), this pulsar shows up as a steep spectrum compact source with a spectral index of −2.2. The integrated flux of the shell, including the pulsar and other foreground or background sources superimposed across the shell, is 1.0 Jy. The rms noise in the image is 10 μJy beam⁻¹.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The technical goals of this pilot project were to assess our low-frequency wide-band imaging capabilities for targets in the Galactic plane. In the L band and C band, the Galaxy is replete with compact and diffuse emission of both thermal and non-thermal origin. Sources often have a mixture of thermal and non-thermal emission, resulting in spectral indices that vary with position across a given source and with spatial scale. Conventional imaging algorithms that rely on multiple narrowband images to derive spectral-index information are limited to the angular resolution allowed by the lowest measured frequency and the single-channel sensitivity. The MS-MFS algorithm was designed to optimally use the multi-frequency measurements to perform spectral-index mapping across diffused sources at the highest sensitivity and angular resolution offered by the data.

The following is a summary of our current status, based on the imaging results shown in the previous section. Our targets were chosen based on existing information about their spatial and spectral structure, in order to test the analysis methods in preparation for a Galactic plane survey.

• 1.  
  Separation of thermal versus non-thermal and steep versus steeper spectra: we see clear morphological separations between regions expected to have different spectral structure from the surroundings. These differences range from ▵α ≈ 0.6 between thermal and non-thermal regions, down to ▵α ≈ 0.2 between structures thought to be cores and shells. In regions with high signal-to-noise ratios (S/Ns; >100), the estimated uncertainty in the spectral index is ±0.03. In regions with low S/N (<100), the uncertainty in the spectral index rises rapidly, and all the images presented here used a Stokes-I S/N threshold of 100.
• 2.  
  Image-reconstruction errors: errors in the spectral reconstructions are dominated by multi-scale deconvolution artifacts that decrease with appropriate choices of multi-scale imaging parameters (a set of scale sizes with which to model the spatial structure). Error estimates from Monte Carlo simulations (by varying the choice of spatial scales) range from ±0.02 when scales are chosen appropriately, up to 0.2 in the extreme case where diffuse emission is modeled by a series of δ-functions. These simulations were performed for the EVLA D-configuration and the errors depend on the precise uv-coverage just as they depend on the standard imaging algorithms. The MS-MFS algorithm reconstructs the frequency dependence of the flux by a polynomial fit across the frequency range on a per component basis. The errors on the estimated spectral index therefore also depend on the errors of the fitted coefficients. These errors depend on variations in S/N (across the frequency range or due to the scale of the emission) in the same way as the errors on the coefficients of any least-squares polynomial fitting algorithm (see Press et al. 2007; Rau & Cornwell 2011 for further details). Further, at the largest spatial scales sampled by the interferometer (which, of course, is telescope dependent), the spectrum is unconstrained by the data and the MS-MFS model does not provide adequate additional constraints. Out of our examples, G55.7+3.4 and G16.7+0.1 contain structure at spatial scales that fall within the unsampled central region of the interferometer sampling function, leading to overall systematic errors in the spectral index across the source. These errors are limited to the largest scales, but in the current implementation of the algorithm, it is not possible to analyze the scales separately. In our examples, G55.7+3.4 is significantly affected by this error and spectra of only the most compact emission can be trusted. Finally, in addition to errors in the spectral index, off-source artifacts are present in the Stokes-I images beyond dynamic ranges of a few thousand (measured as the ratio of the peak brightness to the peak residual near the source). Dynamic ranges for the images presented here range from 2000 for G16.7+0.1 to 10,000 for G19.6−0.2. In most cases, wide-band self-calibration was required to achieve these limits.
• 3.  
  Wide-field wide-band sensitivity of the instrument and related errors: across a 2:1 bandwidth, the size of the antenna power pattern changes enough to produce an average beam that is significantly different from the beams at individual frequencies. The most prominent difference is the near lack of a first null, with sensitivity at the few percent level out to almost three times the half-power beam width (HPBW). The G55.7+3.4 field is an example of this wide-field sensitivity pattern (the reference-frequency primary beam has an HPBW of 30'). For sources far from the phase center, the w-term causes the dominant wide-field error, and these images (in particular G55.7+3.4) are a result of W-Projection combined with MS-MFS. The next dominant error (in Stokes-I and spectral index) is due to the frequency dependence of the primary beam. All the spectral-index maps shown here have been corrected for this effect using the time-averaged PB model, and can be trusted out to the 25% point of the reference-frequency primary beam. The time variability of the primary beams has not been accounted for, and is thought to be the cause of the next dominant error (visible as artifacts around bright sources about 1° from the phase center in G55.7+3.4).

The next steps in our analysis include combining multi-configuration and multi-band data for the G16.7+0.1 and G19.6−0.2 fields to introduce more constraints during image reconstruction, to complete the software integration required to combine MS-MFS and W-Projection with the A-Projection algorithm to correct for time-varying primary beams, and further tests with mosaicking observations as well as tests of how the additional information aids the image reconstruction. These steps are in accordance with our long-term goal of conducting a wide-band mosaicked survey of the Galactic plane at low frequencies.

We used the ATNF Pulsar Catalogue (Manchester et al. 2005), accessible via the URL http://www.atnf.csiro.au/people/pulsar/psrcat/, for the coordinates of the pulsar in the G55.7+3.4 field.