FERMI-LAT AND WMAP OBSERVATIONS OF THE PUPPIS A SUPERNOVA REMNANT

The LAT detects γ-ray photons by conversion into electron–positron pairs in the energy range between 20 MeV to higher than 300 GeV, as described by Atwood et al. (2009). It contains a high-resolution converter/tracker (for direction measurement of the incident γ-rays), a CsI(Tl) crystal calorimeter (for energy measurement), and an anti-coincidence detector to identify the background of charged particles. The LAT has a large effective area (∼8000 cm² on-axis above 1 GeV), a wide field of view (∼2.4 sr), and good angular resolution (∼06 radius for 68% containment at 1 GeV for events converting in the front section of the tracker). The on-orbit calibration is described in Abdo et al. (2009b).

The following analysis was performed using 36 months of data collected from 2008 August 4 to 2011 August 20 within a 15° × 15° region around the position of Puppis A. Only events with Earth zenith angles smaller than 100° were included to reduce contamination from the Earth limb. We used the P7V6 instrument response functions (IRFs) and selected the "Source" events, which correspond to the best compromise between the number of selected photons and the charged particle residual background for the study of point-like or slightly extended sources.

Two different tools were used to perform the spatial and spectral analysis: $\mathtt {gtlike}$ and $\mathtt {pointlike}$. $\mathtt {gtlike}$ is a binned maximum likelihood method (Mattox et al. 1996) implemented in the Science Tools distributed by the Fermi Science Support Center (FSSC).⁸ $\mathtt {pointlike}$ is an alternate binned likelihood technique, optimized for characterizing the extension of a source (unlike $\mathtt {gtlike}$), that was extensively tested against $\mathtt {gtlike}$ (Kerr 2011; Lande et al. 2012). These tools fit a source model to the data along with models for the residual charged particles and diffuse γ-ray emission. In the following analysis, the Galactic diffuse emission is modeled by the standard LAT diffuse emission ring—hybrid model gal_2yearp7v6_v0.fits. The residual background and extragalactic radiation are described by a single isotropic component with a spectral shape described by the file iso_p7v6source.txt. These models are available from the FSSC.

Sources within 15° around Puppis A listed in the Fermi-LAT Second Source Catalog (Nolan et al. 2012, hereafter 2FGL) are included in our spectral-spatial model of the region. We also include extended spatial templates describing the γ-ray emission from RX J0852.0−4622 (Tanaka et al. 2011) and the pulsar wind nebula (PWN) Vela-X (M. Grondin et al. 2012, in preparation). Puppis A is associated with three 2FGL sources within the radio boundary: 2FGL J0821.0−4254, J0823.0−4246, and J0823.4−4305. These sources are either all left free to vary or are replaced by other spatial models. The spectral parameters of sources closer than 5° to Puppis A are left free, while the parameters of all other sources are fixed at the 2FGL values.

Located at a distance of 3° from the SNR Puppis A, the Vela pulsar is the brightest steady γ-ray source in the sky, from which photons are observed up to 25 GeV (Abdo et al. 2010g). To avoid any bias on the analysis of the SNR due to this bright nearby pulsar, all studies involving events below 3 GeV were performed in the off-pulse window of the Vela pulsar. Gamma-ray photons were phase-folded using a timing solution produced from observations made with the Parkes 64 m radio telescope (Weltevrede et al. 2010). To build this timing solution, 163 times of arrival (TOAs) were used covering the period from the beginning of the mission to 2011 August 20. We fit the radio TOAs to the pulsar rotation frequency and first 10 derivatives. The fit further includes 15 harmonically related sinusoids, using the FITWAVES option in the TEMPO2 package (Hobbs et al. 2006), to flatten the timing noise. The post-fit rms is 119.579 μs, or 0.13% of the pulsar phase. This timing solution will be made available through the FSSC.⁹

The off-pulse phase window was chosen between 0.65 and 1.05, slightly larger than the one used for the analysis of the Vela-X PWN (Abdo et al. 2010b) allowed by the larger angular distance between Puppis A and Vela than between the pulsar and its associated PWN. This off-pulse choice offers good statistics without increasing the systematics due to the contamination from the Vela pulsar.

In the study of the morphology of an extended source, one of the major objectives is to have the best possible angular resolution. Therefore, we restrict our LAT data set to events with energies above 800 MeV. This also reduces the relative level of the Galactic diffuse background owing to its softer spectrum. As discussed above, to ensure that we do not suffer contamination from the Vela pulsar, we only selected the off-pulse photons defined as having phases 0.65–1.05.

We fit models to the data using the maximum likelihood framework. The test statistic (TS) of a source is defined as twice the logarithm of the ratio between the likelihood $\mathcal {L}_1$ obtained by fitting the source model plus background components (including other sources) to the data, and the likelihood $\mathcal {L}_0$ obtained by fitting the background components only, i.e., TS = 2 log($\mathcal {L}_1$/$\mathcal {L}_0$). Figure 1 shows a map of the point-source detection significance, evaluated at each point in the TS map for the region around Puppis A using photons with energies >800 MeV. This skymap contains the TS value for a point source of fixed photon index Γ = 2 at each map location, thus giving a measure of the statistical significance for the detection of a γ-ray source with that spectrum in excess of the background. Emission coincident with Puppis A is detected, but also in two regions outside of its shell. We consider these regions of excess γ-ray emission as likely to belong to background sources not recognized in 2FGL. Due to the longer integration time of our analysis (36 months versus 24 months in the catalog) and the overwhelming brightness of the Vela pulsar in the entire phase interval, the appearance of additional sources in our region of interest is expected. These sources are denoted with the identifiers BckgA and BckgB and will be described below.

Zoom In Zoom Out Reset image size

Puppis A was first established as an extended source above 1 GeV by Lande et al. (2012). Following their procedure, we determined the source extension using $\mathtt {pointlike}$ with a uniform disk hypothesis (compared to the point-source hypothesis) for energies above 800 MeV in the off-pulse of the Vela pulsar and above 5 GeV in the all-phase interval. The results are summarized in Table 1. To quantify the significance of the extension of the SNR Puppis A, we define TS_ext as twice the difference between the log-likelihood of an extended source model and the log-likelihood of a point-like source model. Above 800 MeV, TS_ext = 46 (which converts into a significance of ∼7σ for the source extension), demonstrating that the source is significantly extended with respect to the LAT point-spread function (PSF). The fitted radius is 038 ± 004, in good agreement with the size of the SNR as seen in the radio and X-rays. The extension is still significant above 5 GeV with a slightly smaller size of 032 ± 003. Going from lower to higher energy, the radius of the SNR seems to decrease with a fitted position centered on the eastern side of the remnant but, due to the large uncertainty of the fit with the current statistics, this effect is only marginal. This trend is visible in Figure 2 which presents the LAT TS maps above 800 MeV and above 5 GeV when nearby sources and the extragalactic and Galactic components of the background are included in the background model. The emission above 5 GeV is brightest in the east, near the bright eastern knot, where the SNR shock is interacting with a dense cloud, and similar to the radio and X-ray morphologies.

Zoom In Zoom Out Reset image size

It should be noted that during the fit above 800 MeV, we simultaneously localized the sources BckgA and BckgB assuming the best disk spatial model for the SNR Puppis A, presented in Table 1. We followed the procedure described in Lande et al. (2012). BckgA was fit to position α(J2000) = 12577, δ(J2000) = −4217 with a 68% error radius of 006, while BckgB was fit at α(J2000) = 12814, δ(J2000) = −4339 with a 68% error radius of 005. The addition of sources BckgA and BckgB in the source model results in an improved maximum likelihood above 800 MeV with four additional degrees of freedom. The sources have corresponding significances of ∼4.4 and ∼5.5σ (TS = 27 and 39), respectively, demonstrating that these two sources are significant and distinct from Puppis A. In addition, we will show in Section 2.3 that, in a combined likelihood analysis of the spectrum of Puppis A, the emission from these two sources is much softer than the γ-ray emission from the SNR.

To be more quantitative, we have examined the correspondence of the γ-ray emission from Puppis A with different source morphologies by using $\mathtt {gtlike}$ with assumed multi-frequency templates above 800 MeV. For this exercise we compared the TS of the best-fit uniform disk model provided by $\mathtt {pointlike}$ (see Table 1) with TS derived when using morphological templates from ROSAT and the Very Large Array (VLA; 1.4 GHz radio images) assuming a power-law spectrum. The resulting TS values obtained from our maximum likelihood fitting are summarized in Table 2. The radio template, the X-ray template, and the uniform disk all produce improvements from the 2FGL three-point source model, while having fewer degrees of freedom. Finally, we divided the ROSAT and uniform disk templates into two half-disks along the north/south line in celestial coordinates, as indicated in Figure 1, to verify the evidence from $\mathtt {pointlike}$ of an energy-dependent shape. The two half-disk model provides a marginal improvement (∼2.4σ level), in the same trend described above. However, further data are needed to confirm this effect.

Using the different templates described above, we performed maximum likelihood fits and compared the best-fit parameters in the wide energy range 0.2–100 GeV. To avoid any bias due to the brightness of the Vela pulsar, the spectral fit was performed in the off-pulse phase interval. No evidence for cutoff or break is visible and, in all cases, the Fermi-LAT data are well described by a power law such that the differential photon flux, F(E), is given by

$$\begin{equation} F(E)\,=\,\frac{\it dN}{\it dE} \, = \, F_0\,\left(\frac{E}{E_0}\right)^{-\Gamma }, \end{equation} \tag{ 1 }$$

where F₀ is the differential flux at energy E₀ and Γ is the photon index. The uncertainty contours as a function of energy E, called the butterfly, are defined such that the differential flux satisfies

$$\begin{eqnarray} &&\frac{{\Delta }F^2}{F^2} = \left(\frac{{\Delta }F_0}{F_0}\right)^2 + \hbox{ln}^2\left(\frac{E}{E_0}\right)\Delta \Gamma ^2 - \frac{2}{F_0}\,\hbox{cov}(F_0,\Gamma)\,\hbox{ln}\left(\frac{E}{E_0}\right), \nonumber\\ \end{eqnarray} \tag{ 2 }$$

where cov(F₀, Γ) is the covariance term, returned by the MINUIT minimization and error analysis function called by the Fermi-LAT likelihood analysis tool, $\mathtt {gtlike}$, and ΔF, ΔF₀, and ΔΓ are the statistical uncertainties on F, F₀, and Γ, respectively.

As shown in Table 3, the obtained fluxes and spectral indexes are almost the same for all spatial templates used. These spectral fits take into account the γ-ray emission of the two additional background sources BckgA and BckgB. Their positions and spectral parameters (fitted using the uniform disk D1 to describe Puppis A) are given in Table 4.

We searched for spectral variations across the γ-ray emission associated with Puppis A. We allowed an independent normalization and spectral index for the two X-ray halves. Interestingly, the spectrum of the western half-disk is found to be steeper by an index of 0.3 (2σ) above 200 MeV, with respect to the eastern half-disk. However, more statistics are needed to confirm spectral differences between the eastern and western regions. The spectral parameters for each of the two half-disks of the remnant are reported for reference in Table 3. Note that the TS of each individual half is computed by removing only that half, so the total likelihood improvement from the two X-ray halves model (Table 2) is more than the sum of the TS of the individual halves (Table 3).

To ensure that the slightly softer spectrum in the western half-disk is not due to contamination from the compact object inside the SNR Puppis A, PSR J0821−4300, we performed a timing analysis using the same Fermi-LAT data set. The low TS of this side of the remnant combined with the need to keep only photons in the Vela pulsar off-pulse make a blind pulsation search unfeasible. We used ephemerides from Gotthelf & Halpern (2009) with the assumption of a zero-period derivative (Gotthelf et al. 2010) to fold the photon arrival times using the Fermi plug-in distributed with the TEMPO2 software. We applied the H-test periodicity test (de Jager & Büsching 2010) for several cuts in both energy and photon distance from the pulsar location and found no evidence of pulsation with a significance better than H-test = 13.5 (2.8σ). This confirms that with a small spin-down luminosity and a relatively weak magnetic field (Gotthelf et al. 2010) PSR J0821−4300 is not a good candidate for γ-ray emission.

The Fermi-LAT spectral points for Puppis A were obtained by dividing the 200 MeV–100 GeV range into seven logarithmically spaced energy bins and performing a maximum likelihood spectral analysis in each interval, assuming a power-law spectral shape with fixed photon index Γ = 2 and the X-ray spatial model for Puppis A. Spectral points of Puppis A above 3 GeV were determined using the all-phase interval which is allowed by the improved PSF at those energies as well as the lower energy flux of the Vela pulsar. These spectral energy points are in agreement with those derived in the off-pulse window. The result and the butterfly from the overall spectral fit, renormalized to the total phase interval, are presented in Figure 3 along with the butterflies corresponding to the fit of the two half-disks of the SNR. The errors on the spectral points represent the statistical and systematic uncertainties added in quadrature. Two main systematic errors have been taken into account: imperfect modeling of the Galactic diffuse emission and uncertainties in the effective area calibration. The first one was estimated by changing artificially the normalization of the Galactic diffuse model by ±6% as done in Abdo et al. (2010g). The second one is estimated by using modified IRFs whose effective areas bracket the nominal ones. These bracketing IRFs are defined by envelopes above and below the nominal energy dependence of the effective area by linearly connecting differences of (10%, 5%, and 20%) at log₁₀(E/MeV) of (2, 2.75, and 4), respectively.

Zoom In Zoom Out Reset image size

Using the best-fit spatial and spectral model (D1) renormalized to the all-phase interval, the γ-ray luminosity of Puppis A in the 1–100 GeV band is calculated to be 2.7 × 10³⁴ (D/2.2 kpc)² erg s⁻¹. This is among the faintest reported SNRs detected by Fermi-LAT. By comparison, SNRs which are interacting with dense clouds, such as IC 443 and W51C have luminosities of ⩾10³⁵ erg s⁻¹ (Abdo et al. 2009a, 2010f). SNR W49B is thought to have an age of ∼4 kyr, comparable to Puppis A, yet has a luminosity of 9.3 × 10³⁵ erg s⁻¹ above 1 GeV and is interacting with a dense molecular cloud (Abdo et al. 2010d). The relatively low GeV luminosity of Puppis A may be due to either less gas in the vicinity of the SNR, or to an emission mechanism different than in high-luminosity SNRs. We explore this further in Section 3 by fitting simple nonthermal emission models.

The radio morphology of Puppis A is that of a bright shell which greatly overlaps but is not entirely correlated with the X-ray and IR morphologies. The integrated radio spectrum between 19 MHz and 8.4 GHz is well fit by a single power law with slope α = −0.52 ± 0.03 (Castelletti et al. 2006). The high-resolution VLA observations at 1425 and 327 MHz permitted a detailed study of the radio morphology and spectral index of Puppis A (Castelletti et al. 2006). While it is difficult to quantify, spectral flattening, to an index α ∼ −0.4, is observed for the radio-bright regions where interaction with denser gas is thought to take place. In contrast, the periphery of the SNR shell appears to show a steeper spectral index of α ∼ −0.6. The resulting global spectral index is therefore an average of these regions.

We have used the 7 year all-sky data of the WMAP to extend the radio spectrum of Puppis A to higher frequencies. Five bands are analyzed with effective central frequencies (ν_eff) of 23–93 GHz (Jarosik et al. 2011). To obtain WMAP flux densities, we used template fitting of the flux-corrected 1425 MHz radio image (Castelletti et al. 2006). This template image was smoothed with the WMAP beam profiles, and then fit in each band with a constant times, the smoothed template plus a sloping planar baseline. To avoid emission from nearby Vela-X PWN, the template fit is restricted to a circular region within a 2° radius. The 7 year skymaps, fit template, and residuals are presented in Figure 4. Fluxes are given in Table 5 with errors estimated from the rms of the fit residuals in each band. For each band, we list the effective Gaussian beam full width at half-maximum (FWHM). At 23 GHz (K band) the beam width is comparable to the angular diameter of the SNR. However, at higher frequencies, the remnant is resolved as a spatially extended source. There are no previous radio detections of Puppis A at such high frequencies.

Zoom In Zoom Out Reset image size

The full radio spectrum of Puppis A from 19 MHz to 93 GHz is shown in Figure 5. To fit the radio spectrum, we exclude data below 300 MHz, which may suffer from low-frequency absorption by thermal electrons along the line of sight. We find a best-fit 1 GHz flux density of 141 ± 4 Jy and a radio spectral index of −0.56 ± 0.01. A radio index of −0.55 is equivalent to a nonthermal particle distribution index of 2.1, matching the spectral index obtained with the Fermi-LAT data in Section 2.3.

Zoom In Zoom Out Reset image size

The WMAP data above 40 GHz appear to show a decreasing flux, indicative of a spectral break or cutoff. Both the measured flux density at 63 GHz and the 2σ upper limit at 93 GHz fall below the best-fit radio spectrum. WMAP has excellent interband calibration (Weiland et al. 2011), so this offers intriguing evidence for spectral curvature at high radio frequencies. To assess the statistical significance of the putative spectral break, we assume a synchrotron cooling break by Δα = −0.5 and apply the F-test to compare the χ² fit to that of a simple power law. We find a 2.8σ significance for a break at 40 GHz, which is not enough to claim a detection, but highly suggestive. Figure 5 shows that the fit with a spectral break at ν_b ∼ 40 GHz can reproduce the observed data. This is in contrast to the LAT spectrum, where there is no evidence for a break or cutoff at high energies. We discuss the implications of a high-frequency radio break in Section 4.

Puppis A has several photon fields which can be upscattered by relativistic electrons to produce IC emission: the cosmic microwave background radiation (CMBR), the infrared continuum produced by the collisional excitation of swept-up dust, and the local interstellar radiation field (ISRF). The luminosity of IC γ-rays is proportional to n_ph ^1/2, where n_ph is the average photon density and is the average photon energy (Gaisser et al. 1998). Typically, the CMBR is the dominant radiation field, characterized by n_ph ≈ 400 cm⁻³ and ≈ 6.3 × 10⁻⁴ eV. Dust emission from Puppis A is fit by a two temperature blackbody with T₁ = 150 K, n_ph = 12 cm⁻³, and = 0.04 eV; and T₂ = 45 K, n_ph = 20 cm⁻³, and = 0.01 eV (Arendt et al. 1991). Summing these components, the total contribution to IC γ-rays from IR dust emission is only ∼60% that from the CMBR. As Puppis A lies at a Galactocentric radius comparable to the Sun, we assume the local ISRF spectrum (Mathis et al. 1983) which peaks at 1.2 eV and contributes ∼25% compared to the CMBR. While shocked dust and the stellar optical background contribute significantly to IC emission, they do not dominate over the CMBR and cannot solely explain emission from Puppis A.

Accounting for these photon fields, an IC-dominated model with a particle index of 2.1 and a cutoff at 500 GeV provides a good fit to the data. With a magnetic field of ∼8 μG, this scenario requires a total energy in CR electrons of 2.9 × 10⁴⁹ erg. It should be noted that the electron-to-proton ratio is not well constrained and was fixed here at η_e/η_p = 1. However, this value needs to be larger than 0.1, and thus in excess of the ratio found for local CR abundances, to inject a reasonable energy content in radiating electrons. In the same way, the average gas density needs to be lower than 0.3 cm⁻³ to reduce the bremsstrahlung component. This limit is very close to the lower limit from thermal X-ray observations (Hwang et al. 2005) and is somewhat low for unshocked ISM in the vicinity of a molecular cloud.

Bremsstrahlung emission, from CR electrons encountering gas of moderate density, can provide a reasonable emission mechanism. X-ray and IR observations constrain the density in the shock interaction region to be ∼4 cm⁻³ (Arendt et al. 2010). Assuming an electron-to-proton ratio of 1 and a gas density of 4 cm⁻³, the best-fit bremsstrahlung-dominated model has a particle index of 2.1 and a cutoff above 500 GeV. The total energy in CR electrons is 1.3 × 10⁴⁹ erg, and the magnetic field is ∼13 μG. Given that the SNR is adiabatically expanding, compressing the swept-up gas and magnetic field lines by a factor of ∼4, a magnetic field of 13 μG is reasonable for compression alone in the post-shock region from which radio and γ-ray emission may arise. We also note that at densities of ∼4 cm⁻³, bremsstrahlung emission dominates over hadronic π⁰ decay for electron-to-proton ratios greater than ∼0.1, again in excess of CR abundances.

We then consider a π⁰-decay model to account for the broadband spectrum of Puppis A. The flux and spectrum of the γ-rays produced by π⁰-decay are calculated following the analytical approximations by Kelner et al. (2006). Here, we assume a gas density of 4 cm⁻³, and an electron-to-proton ratio of 0.02, consistent with the locally observed CR composition. The particle index is fixed at 2.1 with a cutoff above 800 GeV. The total energy in CR hadrons is 4.0 × 10⁴⁹ erg and for electrons is 2.8 × 10⁴⁸ erg. Assuming a typical explosion energy for Puppis A, this hadronic model indicates that a few percent of the initial kinetic energy is transferred to CRs within a few thousand years. The magnetic field value required to reproduce the synchrotron radio observations is ∼35 μG, assuming that the radio and γ-ray emissions arise from the same volume. However, if the γ-ray emission originates from higher density gas with a lower volume filling fraction than the radio emission, the average magnetic field would be less than our fitted value.

From the three emission models above, it is clear that leptonic scenarios require an electron-to-proton ratio at least 10 times larger than the standard CR abundance ratio. In addition, the IC-dominated model requires an extremely low density as well as a relatively low magnetic field value. However, with the current statistics, it is not possible to discriminate between these different models, and the parameters listed in Table 6 provide an estimate of the energetic particle population in the SNR Puppis A under reasonable assumptions. It should be noted that the CR content (electrons and protons) estimated in the leptonic models should be taken as a lower limit since the electron-to-proton ratio is fixed at a maximum value of η_e/η_p = 1 in the fit.