Asymmetric Expansion of the Youngest Galactic Supernova Remnant G1.9+0.3

The youngest Galactic supernova remnant (SNR) G1.9+0.3 (Reynolds et al. 2008) has provided important new information on the very early development of an SNR as it interacts with the immediate SN environment. Its rapid expansion (∼13,000 km s⁻¹, for a presumed distance of 8.5 kpc; Carlton et al. 2011, hereafter C11) allows for the study of expansion over relatively short time baselines. In C11, we compared observations in 2007 (Epoch I) and 2009 (Epoch II) in X-rays to obtain a mean expansion rate of 0.642% ± 0.049% yr⁻¹, or a mean expansion age of 156 ± 11 yr. Assuming a mean deceleration rate m ($R\propto {t}^{m}$) of about m = 0.7, this implies a remnant age of 100 yr. However, the expansion is far from uniform. A much longer (1 Ms) observation in 2011 (Epoch III) allowed the detection of departures from uniformity in expansion along the major axis (Figure 1): the expansion varies systematically with radius, decreasing from $0.84 \% \pm 0.06 \%$ yr⁻¹ to $0.52 \% \pm 0.03 \%$ yr⁻¹ (Borkowski et al. 2014, hereafter B14). This difference was ascribed either to asymmetry in the surrounding medium, or in the ejecta themselves, indicating a significantly anisotropic explosion. We have now obtained an additional observation in X-rays of 400 ks (2015), which allows further study of nonhomologous expansion in G1.9+0.3. These observations should inform ongoing work on modeling G1.9+0.3 (e.g., Tsebrenko & Soker 2015a, 2015b; Chakraborti et al. 2016; Yang et al. 2016).

Zoom In Zoom Out Reset image size

The high expansion velocities, absence of an obvious pulsar-wind nebula, and bilateral symmetry of the X-ray emission (as in SN 1006; Winkler et al. 2014) suggest that G1.9+0.3 originated in a Type Ia event. The detailed dynamics can then provide clues to the nature of these explosions, at an age when ejecta still dominate the dynamics. Thermal emission from Si, S, and Fe in isolated regions has allowed a picture to be drawn of highly anisotropic, overturned material (Borkowski et al. 2013).

One mystery, evident from the earliest X-ray observations, was the difference in morphology between radio and X-ray images. The radio image (Figure 1) shows a broad maximum across the northern rim, rather than bilaterally symmetric rims in the southeast and northwest. Since all the emission is synchrotron, this morphological contrast is puzzling. SN 1006, for instance, has virtually identical images in radio and nonthermal X-rays (Winkler et al. 2014). The simplest explanation involves assuming that radio-bright but X-ray faint regions result from an energy distribution of relativistic electrons that is larger than that from X-ray bright regions at radio-emitting energies of order 1 GeV, but which cuts off before reaching X-ray emitting energies. One must then search for a mechanism to produce such differences.

2.1. X-Ray

2.2. Radio

Relatively simple, parametric methods for measuring expansion (such as those used in C11 and B14) cannot describe the very complex expansion of G1.9+0.3 revealed by the new Chandra observations. We now use the "Demons" algorithm of Thirion (1998), as implemented in the SimpleITK software package (Lowekamp et al. 2013), to measure the nonuniform expansion of G1.9+0.3. Attributing morphological changes only to motions, the Demons method provides a nonparametric way for measuring these motions globally. Since this computationally efficient method (approximately) minimizes the sum of squared differences between pixel intensities, we need to apply it to the smoothed images, not to raw Chandra images that are badly affected by counting noise.

We smoothed Chandra data with the non-local PCA method of Salmon et al. (2014), with smoothed images extracted from smoothed data cubes. This method combines elements of dictionary learning and sparse patch-based representation of images (or spectral data cubes) for photon-limited data. Because this Poisson-PCA method is computationally intensive, relatively small (${512}^{2}\times 15$) data cubes, heavily binned along the spectral dimension as described above, were smoothed using patches ${10}^{2}\times 6$ in size. The moderate spatial patch size of $2\buildrel{\prime\prime}\over{.} 46\times 2\buildrel{\prime\prime}\over{.} 46$ preserves sharp spatial structures seen in the bright filamentary features, while a large patch size in the spectral dimension is suitable for the synchrotron-dominated spectra of G1.9+0.3 that vary smoothly across the entire spectral range of Chandra. With the patch size chosen, the most important parameters that control the smoothing of data cubes are the rank l of the approximation to the underlying intensity to the collection of patches (order of the Poisson-PCA method), and the number of clusters K into which patches are grouped prior to estimation of intensities. We used l = 6 and K = 30 for the 2011 and 2015 data cubes of G1.9+0.3.

The expansion of G1.9+0.3 is readily apparent by comparing smoothed Chandra images from 2011 and 2015 (Figure 2). We measure an average expansion rate of 0.58% yr⁻¹ between these two epochs (using the same method as in C11), but there are very large deviations from uniform expansion. In only four years, striking morphological variations at various locations within the remnant can be discerned by eye. These include bending of the innermost rim in the west, and even more complex, strongly nonradial motions in the northeast where changes in relative positions of prominent emission knots and filaments can be seen. In the framework of the Demons algorithm, which can account for such complex motions, the 2011 1–8 keV smoothed image is considered as the reference image, while the 2015 1–8 keV smoothed image is the moving image. Before application of the Demons method, a particle background with an estimated rate of $9.0\times {10}^{-5}$ and $7.0\times {10}^{-5}$ cts s⁻¹ per image pixel was subtracted from the 2011 and 2015 images, respectively, and then each background-subtracted image was normalized using a corresponding monochromatic (E = 3 keV) exposure map. Point sources and the featureless interior of the remnant have been masked out. The uniform expansion at 0.58% yr⁻¹ centered $10\buildrel{\prime\prime}\over{.} 5$ north of the geometrical center was used to set initial displacements between the two epochs, then at each iteration step of the Demons method computed displacements were smoothed with a Gaussian with FWHM of $0\buildrel{\prime\prime}\over{.} 82$.⁶ After achieving convergence, this smoothing resulted in a good match between the 2011 image and the (transformed) 2015 image (Figure 3), demonstrating the effectiveness of the Demons method, but the final displacement vectors were affected by substantial noise. We reduced this noise by calculating brightness-weighted averages of displacement vectors within various regions of interest. Finally, these averaged displacements were divided by ${\rm{\Delta }}t$ to obtain proper motion vectors ${\boldsymbol{\mu }}$.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We show vectors ${\boldsymbol{\mu }}$ in Figure 3 for a large number of regions chosen to delineate all major morphological features of G1.9+0.3. The white arrow with length $0\buildrel{\prime\prime}\over{.} 25$ yr⁻¹ (10,100 km s⁻¹ at 8.5 kpc) shows their scale, while their tails are located at the geometrical centroids of these regions. Motions are between 009 yr⁻¹ and $0\buildrel{\prime\prime}\over{.} 44$ yr⁻¹, varying by a factor of up to 5. The median μ is $0\buildrel{\prime\prime}\over{.} 28$ yr⁻¹. The slowest shocks are in the north. The fastest motions are predominantly along the major axis of the remnant. Motions are strongly nonradial with respect to the remnant's geometrical center at R.A. 17^h48^m4561, decl. −27°10'056. We infer the explosion occurred somewhere northeast of the geometrical center. Extreme deviations from radial motions are present.

In Figure 4, we draw vectors ${\boldsymbol{\mu }}$ over the radio and contemporaneous 2009 Chandra images (tails of vectors were shifted inward to account for the remnant's expansion between the 2008 December array A VLA and Epoch III Chandra observations). Expansion is noticeably slower along the outer boundary of the radio emission. The median motion along the outer radio contour at 1.5 mJy beam⁻¹ (regions a to n in Figure 4) is $0\buildrel{\prime\prime}\over{.} 17$ yr⁻¹ (only 6900 km s⁻¹), significantly less than the overall median of $0\buildrel{\prime\prime}\over{.} 28$ yr⁻¹ (11,000 km s⁻¹). Among all regions with $\mu \lt 0\buildrel{\prime\prime}\over{.} 17$ yr⁻¹, most (12 out of 14) are within the northern rim, predominantly toward the north and northeast where radio emission is particularly bright. This includes regions f ($0\buildrel{\prime\prime}\over{.} 16$ yr⁻¹), g ($0\buildrel{\prime\prime}\over{.} 13$ yr⁻¹), h ($0\buildrel{\prime\prime}\over{.} 14$ yr⁻¹), i ($0\buildrel{\prime\prime}\over{.} 095$ yr⁻¹), k ($0\buildrel{\prime\prime}\over{.} 11$ yr⁻¹), l ($0\buildrel{\prime\prime}\over{.} 089$ yr⁻¹), and n (013 yr⁻¹). The smallest μ (region l) corresponds to a velocity of only 3600 km s⁻¹. But motions of several knots and filaments in the northeast are larger, more typical for the remnant as a whole, including regions j ($0\buildrel{\prime\prime}\over{.} 26$ yr⁻¹) and m ($0\buildrel{\prime\prime}\over{.} 29$ yr⁻¹). The latter corresponds to a very bright X-ray knot with motion strongly deviating from radial (by 42°), while the former is the northernmost of five fast-moving knots/filaments that are mostly located near where G1.9+0.3 is the brightest in radio.

Zoom In Zoom Out Reset image size

The motions along the northwest outer radio contour (regions a − e) are $0\buildrel{\prime\prime}\over{.} 20$ yr⁻¹, $0\buildrel{\prime\prime}\over{.} 27$ yr⁻¹, $0\buildrel{\prime\prime}\over{.} 29$ yr⁻¹, $0\buildrel{\prime\prime}\over{.} 19$ yr⁻¹, and $0\buildrel{\prime\prime}\over{.} 24$ yr⁻¹, respectively, with a median of $0\buildrel{\prime\prime}\over{.} 24$ yr⁻¹ (9700 km s⁻¹). This is significantly more than those in the north−northeast, where the median for regions f − n is $0\buildrel{\prime\prime}\over{.} 13$ yr⁻¹ (5300 km s⁻¹), but still less than average for G1.9+0.3. As in the northeast, strongly nonradial motions are present. Unlike the north−northeast where X-rays are generally faint and radio is quite bright, this region is moderately bright both in X-rays and radio.

The motions along the major axis are faster than the overall median of $0\buildrel{\prime\prime}\over{.} 28$ yr⁻¹. For the brightest X-ray rims in the middle, the median is $0\buildrel{\prime\prime}\over{.} 32$ yr⁻¹, increasing modestly to $0\buildrel{\prime\prime}\over{.} 33$ yr⁻¹ ahead of these rims. For the innermost rims (this includes vectors ${\boldsymbol{\mu }}$ within the inner pair of regions in B14), the median drops to $0\buildrel{\prime\prime}\over{.} 30$ yr⁻¹. These rather modest (12,000—13,000 km s⁻¹) radial variations along the major axis are not surprising in view of the rapid radial decrease in expansion rates found by B14.

For the southern radio rim, the motion is $0\buildrel{\prime\prime}\over{.} 21$ yr⁻¹ (the median for 7 regions along this rim). This is significantly more than the $0\buildrel{\prime\prime}\over{.} 13$ yr⁻¹ that we found in the north−northeast, and corresponds to velocity of 8300 km s⁻¹. The southernmost extension toward the east, with only a very faint radio counterpart, must be moving much faster than this, as we find $\mu =0\buildrel{\prime\prime}\over{.} 38$ yr⁻¹ (15,000 km s⁻¹). Apparently, the motions are also complex in this relatively faint region of the remnant.

A joint fit to the spatially integrated 1–9 keV spectrum of G1.9+0.3 and to the background spectrum gave a best-fit photon index Γ of 2.45 ± 0.06 and an absorbing column density ${N}_{{\rm{H}}}=7.41(\pm 0.16)\times {10}^{22}$ cm⁻², with an absorbed 1–7 keV flux ${F}_{2015}$ of $3.063(\pm 0.027)\times {10}^{-12}$ erg cm⁻² s⁻¹ (errors are 90% confidence intervals). Both Γ and N_H are consistent with previous measurements (${\rm{\Gamma }}=2.40\pm 0.03$ and ${N}_{{\rm{H}}}=7.25(\pm 0.09)\times {10}^{22}$ cm⁻²; B14), while the flux increased by 3% since 2011. So G1.9+0.3 continues to increase in brightness, but at a somewhat lower rate than the $1.9 \%$ yr⁻¹ reported in B14.

A linear regression fit to ${F}_{2015}$ and to the previously measured fluxes (${F}_{2007}=2.73\times {10}^{-12}$ erg cm⁻² s⁻¹, ${F}_{2009}=2.88\times {10}^{-12}$ erg cm⁻² s⁻¹, ${F}_{2011}=2.97\times {10}^{-12}$ erg cm⁻² s⁻¹; B14) gives a rate of $1.3 \% \pm 0.8 \%$ yr⁻¹. Flux residuals (model fit minus observations) range from $-1.4 \%$ (2011) to $1.2 \%$ (2007), comparable to the ($1\sigma$) statistical error of $1.3 \%$ for ${F}_{2007}$ but more than the 0.3%–0.6% found at other epochs. Although only about 8000 source counts were detected in the short 2007 observation, systematic errors are already becoming comparable to statistical errors for Chandra spectra with that number of counts ($\sim {10}^{4}$). The much longer observations from later epochs are then dominated by systematic (not statistical) errors, and the errors are likely comparable for all four epochs. In this case, the flux residuals quoted above are consistent with a linear flux increase from 2007 to 2015. Our newly determined, more accurate rate of $1.3 \%$ yr⁻¹, with the relatively large error of $\pm 0.8 \%$ yr⁻¹ accounting for systematic effects, is also consistent with the larger rate of $1.9 \%$ yr⁻¹ ($\pm 0.4 \%$ yr⁻¹—statistical errors only) based on the Epoch I–III observations alone (B14).

Figure 4 shows that the slowest expansion speeds are found just beyond the radio maxima. Evidently the shock is encountering denser material there, resulting in a drastic deceleration of the expansion. Such dense material in a slower shock will have two effects on the accelerated electron spectrum: a larger number density of accelerated electrons (since that density is generally found to be a fraction $\sim {10}^{-4}$ of the total particle density; e.g., Ellison & Cassam-Chenaï 2005), and a lower maximum energy, since ${E}_{\max }\propto {{Bv}}_{\mathrm{shock}}^{2}$ for age-limited acceleration (e.g., Reynolds 2008). The resulting synchrotron emission will be brighter in radio than that from surrounding regions with faster shocks, but the spectrum will turn over at lower energies and may not reach to X-rays. That is, the slower shocks we observe, produced by denser ambient material, can explain the differing radio and X-ray morphologies. Assume (conservatively) $B\propto {\rho }^{1/2}$ (no extra magnetic-field amplification). Then, since we expect $\rho {v}_{\mathrm{shock}}^{2}\sim$ const., we have ${E}_{\max }$ (age) $\propto {\rho }^{-1/2}\propto {v}_{\mathrm{shock}}$, and ${\nu }_{\mathrm{roll}}\propto {E}_{\max }^{2}B\propto {v}_{\mathrm{shock}}$. In the region of the radio maximum, the mean shock speed is about 4000 km s⁻¹, or about 0.3 of the value at the bright rims. The integrated spectrum of G1.9+0.3 is well described by a rolloff of $3.1\times {10}^{17}$ Hz (Zoglauer et al. 2015), implying a rolloff near the radio peak of about $9\times {10}^{16}$ Hz or less than 0.4 keV—too low to produce appreciable X-ray synchrotron emission at that location.

The rate of flux increase we find, 1.3% yr⁻¹, is still in the range of radio increase rates of 1%–2% yr⁻¹ (Green et al. 2008; Murphy et al. 2008), so the important question of the comparison of radio and X-ray rates is still unsettled. Different rates imply time-variation in the maximum electron energies, or something even more unexpected. Further observations of both radio and X-rays will be crucial.

The large range of velocity vectors, in both magnitude and direction, that we find in G1.9+0.3 highlights the inadequacy of spherically symmetric models for describing SN events and their aftermath. The north rim is particularly poorly described by 1D models. Failure to account for the large variations in velocity casts doubt on such analyses. However, these velocities contain a great deal of information on the explosion and the circumstellar environment that rivals Kepler's SN, the most recent historical SN in the Galaxy, in its complexity. As in Kepler's Type Ia SNR, large ($\gt 10$) density gradients are present, and SN ejecta are likely colliding with the asymmetric circumstellar medium ejected by the SN progenitor. Our statistical method of determining velocities throughout an expanding remnant, and not just in well-defined sharp outer edges, should have wide application in following the expansion of young SNRs and elucidating their early evolution and the nature of the surrounding medium.

We acknowledge support by NASA through Chandra General Observer Program grants SAO GO5-16069A—C.

Facilities: CXO - Chandra X-ray Observatory satellite, VLA - .