THE SWIFT BURST ALERT TELESCOPE PERSPECTIVE ON NON-THERMAL EMISSION IN HIFLUGCS GALAXY CLUSTERS

For the lower energy BAT bands, it is very useful to have X-ray spectra at lower energies to constrain the thermal emission; this is particularly true given that the Swift BAT survey spectra are coarsely binned (eight channels spanning 14 keV <E < 195 keV). Also, any non-thermal component in the BAT spectra must be consistent with the spectra at softer energies. XMM-Newton is the ideal observatory to provide such complementary spectra. For one, its large field of view (FOV) allows a higher fraction of the total emission, which can be quite extended given the low redshifts of the sample, to be detected in a single pointing. Additionally, the EPIC instruments are sensitive to 5–10 keV photons, which make them more useful for constraining the highest temperature gas, and the telescopes have good spatial resolution so that point sources can be excluded from the spectra. Last, but of no less importance, XMM-Newton has observed all but one (A2244) of the clusters in HIFLUGCS. Unfortunately, another four cluster observations (A401, A478, A1736, and A2163) are heavily contaminated by background flares and consequently unusable (for more details, see Zhang et al. 2011). However, the data for the remaining 59 clusters are of sufficient quality to help constrain potential non-thermal signals in the BAT energy bands.

We extract XMM-Newton spectra for each cluster from the largest circular region that either covers the FOV or extends to the point where cosmic X-ray background (CXB) emission begins to dominate, by summing the annular spectra from Zhang et al. (2009). To ensure near-Gaussian statistics for χ² fitting, adjacent channels are grouped until each new bin contains at least 30 counts. The centers and radii of the circular regions, along with each pointing's observation ID, are listed in Table 1. Source spectra are extracted in concentric annuli within the region; corresponding particle background spectra are derived from CLOSED mode calibration data, which are renormalized based on 3–10 keV events out of the FOV and outside of a 154 radius from the detector center (for details see Section 2.4 of Zhang et al. 2009). The full background treatment is described in Zhang et al. (2009). As an additional step, we readjust the normalization of the particle background spectra by hand to ensure that the 7–12 keV continuum of the cluster spectra has a more physical shape. We define "more physical" as the background normalization that minimizes the χ² statistic for a single temperature (1T; using the APEC plasma emission model⁷) individually fit to the EPIC-pn (2 < E < 12 keV) and MOS1 and MOS2 (2 < E < 10 keV) spectra. The new best-fit temperatures, after these initial renormalizations of the background, are compared to each other and to previous measurements (primarily Reiprich & Böhringer 2002). While this method may bias the background level, especially if a single-temperature model is a poor description of a given spectrum, repeating this procedure with two-temperature (2T) and single-temperature plus power-law (T+NT) models yields comparable or inferior results, usually favoring obvious undersubtractions of the background that produce systematic patterns in the residuals. We favor normalizations that leave the background slightly undersubtracted, in order to avoid removing a real non-thermal signature. For the most part, the overall spectrum is only mildly affected since much of the emission is at lower energies where the background is a smaller fraction of the total. One consequence is that instrumental lines, which are typically between 7.5 and 9.5 keV and are mainly a problem in the EPIC-pn spectra and which can vary in intensity relative to the background continuum, can be under- or oversubtracted. No resolved ICM lines exist in this range, so we simply ignore this energy range when poor line subtractions occur, as in Wik et al. (2009). Based on the change in χ² as the background normalization is varied, a typical 90% level uncertainty in the normalization is ∼3%.

We choose to model, instead of subtract, one further background component: the CXB due to extragalactic sources. Lumb et al. (2002), using XMM-Newton sky fields, find that this component of the CXB is well fit by a power law with a photon index of 1.42 in the hard band (2–10 keV). Their results are in good agreement with other work in this band (e.g., Moretti et al. 2003; De Luca & Molendi 2004). We adopt their normalization at 1 keV of 8.44 photons cm⁻² s⁻¹ keV⁻¹ sr⁻¹, which is scaled to match the extraction area for each cluster. The impact of cosmic variance, or the field-to-field variation in CXB flux resulting from large-scale structure and source population selection, is not included as a systematic uncertainty in the following analysis due to its small effect. While cosmic variance increases with a decreasing solid angle, the high sensitivity of XMM-Newton allows most of the sources responsible for a higher variance to be removed, so for one of our typical regions the 90% uncertainty is only ∼10% of the CXB flux. Note that Lumb et al. (2002) remove detected point sources as is done here, so their spectrum can be directly applied as is. The Galactic component of the CXB is also not considered, as it only contributes below 1 keV, and we restrict our fits to the 2–12 keV range.

The Swift mission and the properties of the survey are described in detail in Wik et al. (2011, their Section 2.2) and in Tueller et al. (2010). Similarly, we refer to that section and the appendices for details on the extraction and calibration of sources from survey image data. To briefly summarize, the flux calibration is tied to the Crab spectrum, which we define to have the same spectrum as that observed by XMM-Newton for E > 2 keV, extrapolated to BAT energies via an adopted model based on Suzaku observations. In this way, both the cross-normalization and spectral shape of the XMM-Newton and Swift spectra will match, and continuous models can be jointly fit to them simultaneously. Unfortunately, independent measurements by various instruments, including the Swift BAT, have recently demonstrated that the hard X-ray spectrum of the Crab is in fact variable on yearly timescales (Wilson-Hodge et al. 2011). At 14–50 keV energies, where the only appreciable amount of flux is detected from clusters, the variation spans about 10% over the last five years, with a consistent decline only over the last two years (see Figure 5 of Wilson-Hodge et al. 2011). Our flux calibration of BAT sources depends on an adopted model for the Crab spectrum, which is taken from Suzaku XIS and HXD-PIN observations that took place in 2005 August; this time occurs during one of the higher flux periods. Since the BAT survey spectrum of the Crab spans the following five years of observations and averages over these fluctuations, the normalization of our adopted model is only 2%–3% higher than the actual flux emitted. The effect this has on our derived fluxes is to make them 2%–3% higher than they actually are; this amount is equivalent to the 1σ error on the 14–20 keV flux of Coma, which is the highest signal-to-noise flux considered here by a factor of two. Also, Wilson-Hodge et al. (2011) show that the recent decline in flux is more dramatic for higher energy bands. Both of these behaviors—the overall decline in flux and the steepening of the spectrum—bias our derived BAT fluxes high, which could lead to a higher chance of false non-thermal detections. However, since we find no convincing evidence for non-thermal excesses even given this probable effect, and since we allow for a 10% cross-calibration uncertainty between the BAT and EPIC spectra, which easily encompasses this level of variability, it is clear that our choice of a flux calibrator does not strongly impact the following analysis, except to make our upper limits slightly more conservative than they otherwise would be.

While the standard processing of coded mask imaging data is designed to extract the fluxes of point sources, it is also possible to extract the flux of a mildly extended source, albeit with somewhat greater uncertainty (Renaud et al. 2006; Wik et al. 2011). The large effective point-spread function (FWHM ∼20') for point sources in the survey means that even nearby clusters of galaxies will appear only slightly extended; the FWHM of the Coma cluster—the most extended, reliably detected source in the survey—is only 285. Note that while four clusters (Fornax, NGC 4636, A3526, and A1060) have larger angular extents than Coma—based on angular R₅₀₀ estimates (Eckert et al. 2011)—they are all cooler, less massive systems and thus either not detected or only marginally detected by the BAT at 14–20 keV. From Figure 1, it is clear that detected clusters (colored circles) are typically extended, relative to other sources. The horizontal lines mark the standard deviation of best-fit FWHM values for the non-cluster sources in each signal-to-noise bin; they also represent the approximate error on FWHM estimates for the clusters in each bin. Individual clusters are labeled in the four lowest energy BAT bands when they are detected at a signal-to-noise ratio (S/N) greater than 5. We follow the procedure outlined in Wik et al. (2011) to extract fluxes for diffuse sources, which requires the spatial distribution of the emission to be known. Because clusters are comparable in size to the effective spatial resolution of the survey, detailed spatial models are not necessary to extract accurate fluxes. We consider generic β-model surface brightness profiles, which well represent the radial profiles at softer energies. Taking a representative value for β of 0.75, we find that all >3σ detected clusters (in a given band) can be well fit with core radii r_c of either 4', 6', 8', or 10'. Profiles with r_c < 4' are hard to distinguish from point-source profiles, so any cluster emission that is too narrow to be fit with the r_c = 4' model is treated as a point source. The true spatial distribution may differ from these fiducial models, but our aim is only to extract accurate fluxes, not to describe the distribution of hard X-ray emission. For Coma, a β-model fit in the first BAT band (E1: 14–20 keV) yields a total flux 9% lower than that derived from a more detailed model of its spatial distribution derived from an XMM-Newton temperature map (see Wik et al. 2011), which accounts for the NE–SW non-axisymmetric elongation of the emission (Eckert et al. 2007). While 9% is a significant difference, Coma is one of the most significantly detected and is the most extended cluster in the survey, so this deviation, which amounts to a factor of only 1.6 times the 1σ error on the flux, is the largest we would expect using this set of extended models. Also, note that no energy dependence in FWHM values is detected; e.g., Coma shows some variation with energy band, but these measurements are all consistent within their uncertainties.

Zoom In Zoom Out Reset image size

We also investigated the use of diffuse models for all the clusters, irrespective of their observed extent, to account for the possibility that we are missing low surface brightness emission obscured by noise. Since the spatial distribution of E > 10 keV emission is unknown, we assume β-model profiles derived from ROSAT images (Reiprich & Böhringer 2002). For clusters with a clearly extended BAT profile, these models reasonably, but usually not perfectly, follow the emission; however, these profiles cannot be reliably distinguished from those at lower energies given that background fluctuations can still distort the profile due to the low S/Ns. Spectral fits using these fluxes produce similar results to those we present in this work, but because their associated errors are larger, these spectra are generally less sensitive, so any additional flux captured—which is not significant—is also diluted. Therefore, these spectra are not considered further.

For clusters with modeled extended emission, we do not want to include the portion of flux that falls outside the XMM-Newton extraction region during joint fits of the data, since the complementary softer flux in the XMM-Newton band spectra is not present. Therefore, only the fraction of the flux that resides within the XMM-Newton region is included in the spectra derived here. One uncertainty, particularly when emission is detected at lower significance, is where the emission is actually coming from, given the positional accuracy of the survey (a 5σ source detected in a given band has a 90% error circle of radius 6'). Since the E1 band-derived positions are near the center of the extraction region, within their respective error circles, we assume that the center of the hard-band distribution is coincident with the center of the XMM-Newton extraction region except for A754, A3266, and A2256. For these detected clusters, their BAT positions are somewhat offset from the surface brightness peak due to an anisotropic temperature distribution produced by mergers (see, e.g., Henry & Briel 1995; Finoguenov et al. 2006; Sun et al. 2002). Following this procedure, we will not underestimate the coincident flux, although overestimates may result that could lead to incorrect hard excesses. However, since we are unable to significantly detect non-thermal emission individually in any of the clusters, this procedure can only cause us to be biased in favor of more conservative upper limits.

The motivation for including XMM-Newton spectra in the analysis is to fully characterize the thermal properties of the hottest gas in the ICM, which will contribute flux to the BAT energy bands. Similarly, these lower energy spectra must be consistent with any indication of a non-thermal component in the BAT spectra; for example, a steep power law may best describe the BAT data but at lower energies result in a poor description of the spectrum. Since our purpose is not to fully characterize the total emission detectable by XMM-Newton, but only to capture the state of the hottest gas, we ignore all events with energies below 2 keV. Cool (≲ 1 keV) gas is completely unimportant at BAT energies, and it will not overly bias E > 2 keV data. We therefore initially consider EPIC spectra in the 2–12 keV range for the pn and 2–10 keV range for the MOS detectors; including photons down to 2 keV provides additional leverage during spectral fitting, since most of the detected photons, regardless of temperature, are at lower energies.

However, the lower end of this energy range presents two issues. First, bright ∼1 keV gas can significantly contribute to the emission between 2 and 3 keV, which certainly exists in some of the cool core clusters in HIFLUGCS. In single-temperature fits, the average temperature will then be biased low to accommodate this component, which could lead to thermal emission being interpreted as a non-thermal excess. Multi-temperature fits would alleviate this problem, but most of the XMM-Newton data are not of sufficient quality to strongly constrain more than one temperature component in this energy range. Including E < 2 keV data to better constrain multi-temperature fits would also require a more complicated analysis that will involve more free parameters and, because the highest S/Ns are in the ∼1 keV channels, fits would be driven by these data, possibly resulting in biased high-temperature components. The second issue relates to the imperfectly calibrated gold edge at 2.2 keV, where the response drops somewhat abruptly. While on its own this feature does not strongly impact spectral fits, because it lies near the edge of our energy range where the S/N is largest, secondary model components can be "co-opted" into better fitting this edge. For instance, in a spectrum truly described by a gas at a single temperature, the addition of a second temperature or non-thermal component to the fit will cause the second component to "fix" any deviations at this edge, typically resulting in a low temperature or steep photon index that has no real physical counterpart.

In practice, both of these effects can conspire to produce the appearance of a more significant non-thermal spectral component than is warranted by the rest of the data. To counter both issues, we also perform fits to data with energies E > 3 keV, which exclude the gold edge and any sizable emission from ≲ 1 keV gas. These spectra have lower S/N due to excluding the 2–3 keV emission, but the high fluxes of clusters in our sample reduce the importance of this issue. Single-temperature fits in both the 2–12 keV and 3–12 keV ranges, jointly fit to all three EPIC spectra (except for A3526, for which the MOS1 spectrum is ignored, and for A2142 and A2147, for which the MOS2 spectra are ignored), are given in Table 2. The pn and MOS instrument cross-normalization is left as a free parameter, which allows for a typical (10 ± 10)% difference between their calibrations (e.g., Snowden 2002). This cross-normalization factor is used and kept fixed during all subsequent joint EPIC-BAT fits. The change in the best-fit temperature from the E > 2 keV to E > 3 keV fits is only ∼0.3 keV on average, indicating that the temperature is generally robust to the choice of the energy range, but that higher energy photons come preferentially from higher temperature gas, assuming that the true temperature structure is not isothermal but contains a continuous spectrum with gas at many temperatures due to substructure and/or radial gradients (Cavagnolo et al. 2008; Snowden et al. 2008).

Our goal is to detect a non-thermal spectral component at hard energies, but because the statistical weight of the BAT channels is so much less than the EPIC channels (lower S/N and fewer of them, at least by an order of magnitude), we have to be careful not to let the XMM-Newton data unfairly drive the spectral fits. To assess the sensitivity of our BAT spectra, we extract 10,000 blank sky spectra from uniformly distributed, random positions at least 40' from any known sources and greater than 20° from the Galactic plane, to mimic the selection function in HIFLUGCS. We then fit these spectra with a fiducial power-law model of photon index Γ fixed at a value of 2, roughly the appropriate slope for IC emission inferred from radio halos, relics, and mini-halos. While the spectral index determined from the radio is typically steeper than this (2.2–2.4), the electrons producing the radio emission at ν > 100 MHz have higher energies than those producing IC at E < 50 keV for B ≲ 0.5 μG, so a simple extrapolation may not be appropriate. A clear flattening of the radio spectrum at low frequencies is apparent in some cases, e.g., Coma (Thierbach et al. 2003) and A3562 (Giacintucci et al. 2005), although this is not universally found as in A2256 (Brentjens 2008) and A2255 (Pizzo & de Bruyn 2009). Since the BAT data are not particularly sensitive to the precise value of the index, we choose a flatter slope to avoid poorly fitting the data at ∼ keV energies where the power-law distribution of relativistic electrons is most likely to turn over in a steady-state-like injection model (e.g., Sarazin 1999).

The distribution of best-fit normalizations from these power-law fits is presented in the narrow histogram in Figure 2. They are well fit by a symmetric Gaussian (dashed smooth line) and indicate a 1σ sensitivity threshold of ∼2 × 10⁻¹² erg cm⁻² s⁻¹ (20–80 keV). Similarly, the formal 3σ detection level is 5.8 × 10⁻¹² erg cm⁻² s⁻¹. In principle, the BAT survey is sensitive enough to confirm or reject previous detections of hard excesses with fluxes ∼10⁻¹¹ erg cm⁻² s⁻¹ (e.g., Rephaeli & Gruber 2002; Molendi et al. 2002; Fusco-Femiano et al. 2004).

Zoom In Zoom Out Reset image size

Now we wish to compare our cluster spectra with this distribution, but first we have to account for any thermal emission in the lower energy bands. The single-temperature models derived with XMM-Newton (2–12 keV) are included as a second component along with the power-law model, with only its normalization left as a free parameter. The resulting non-thermal normalizations are also given in Figure 2 as both the wider histogram (scaled up) and the vertical lines (showing individual values). While the majority of cluster non-thermal components are consistent with the blank sky fits, there is a tail at positive normalizations possibly indicative of a non-thermal excess. However, the thermal contribution is not well determined in this method and may be over- or underestimated. Intriguingly, the three clusters with the most significant non-thermal component (A2029, A1367, and A1651) have positive fluxes, although marginally detected, in all eight BAT bands; this rarely occurs for the blank sky spectra. We discuss these clusters in more detail later. The main result from this analysis is that the BAT cluster spectra have probably not reached a sensitivity level sufficient to detect hard, non-thermal excesses, if they exist, in the brightest clusters.

For each cluster, three simple spectral models are employed to describe the emission covering two orders of magnitude in energy: a single-temperature thermal model (1T), a two-temperature model (2T), and a thermal plus non-thermal model (T+IC). Due to the limited sensitivity of the Swift data, more complicated models cannot be constrained; for example, the separate temperature components in the 2T model are generally poorly constrained in our analysis. Above 50 keV, the APEC emission model is replaced with MeKa because APEC is not defined above 50 keV in the implementation of XSpec used here (version 12.6.0k). Note that the MeKaL emission model could also be used continuously across this energy range, if the look-up table switch is turned off. For the thermal component, the temperature, abundance, redshift, and normalization are all varied. The individual abundances and redshifts in the 2T model are tied together. The non-thermal photon index is initially fixed at Γ = 2, and the normalization is allowed to vary; when the photon index is fit for, it is always fixed to the best-fit value before errors for other parameters are derived. In general, the photon index is poorly constrained, allowing for a wide range of normalizations, which are then less straightforward to evaluate. The purpose of fitting for the photon index is to make sure that we are not biased against detectable IC components with indices that differ from the fiducial value.

Because of complications arising at energies between 2 and 3 keV (see Section 3.1), we perform these fits for both the 2–195 keV (Table 3) and the 3–195 keV (Table 4) spectral ranges. The E > 2 keV fits, at first glance, suggest that there may be evidence for a non-thermal component in a majority of HIFLUGCS clusters. Many of the clusters with some evidence, at least at the 90% level, of a non-thermal excess are, unexpectedly, low-temperature clusters without significant detections at BAT energies. In these cases, the non-thermal component is serving to "adjust" a problem at lower energies—due to either incompletely modeled low-temperature components, an imperfectly calibrated response at the gold edge, or both. The significance of these instances will disappear from fits within a slightly higher energy range, while real non-thermal emission will become a higher proportion of the total flux, and so this component should not greatly diminish in significance. A drastic reduction in the number of marginally detected non-thermal excesses is seen when comparing Tables 3 and 4; only six clusters are seen to have such emission at the 90% confidence level (statistical). These clusters will be discussed individually in Section 4.2.

While the 3–12 keV band avoids some possible systematic uncertainties with the XMM-Newton response and complications from cooler gas, the narrower range may reduce our ability to strongly constrain multi-temperature components in the spectra. One concern is that a weak non-thermal emission component might be indistinguishable from a purely thermal model with a slightly elevated temperature. Note, however, that the 3–12 keV band temperatures in Section 3.1 are typically only ∼0.3 keV higher than the 2–12 keV temperatures. Therefore, the 1T model temperatures should agree for the joint fits over both energy ranges, which is found to be the case in Figure 3. Temperatures derived from joint fits are consistent with those found using only the XMM-Newton spectra, for both energy ranges. For the most part, temperatures from the joint 3–195 keV fits are in good agreement with or slightly lower than the 3–12 keV temperatures. The contribution of the BAT data in this case is to somewhat lower the best-fit temperature, contrary to the expectation if a detectable non-thermal excess were present. The 3–195 keV non-thermal flux limits and possible detections (90%, statistical) are shown in Figure 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Six clusters have a formal detection of non-thermal emission in the 3–195 keV band. Two of these six clusters are also in the top three of candidates for emission based on their BAT-only fits: A1651 and A2142. The other cluster in this top three—with the largest non-thermal normalization of all the clusters—is A2029, so we will include this cluster with the six "detected" clusters as worth some brief discussion. The clusters are listed in the order of decreasing non-thermal flux.

A2029 (Figure 5(a)). This hot (∼8 keV), cool core cluster has been studied in detail with Chandra by Clarke et al. (2004), who explore the interaction between cool gas and the radio active galactic nucleus (AGN) in the cluster center. The cluster is elongated but relatively regular. No evidence exists for major merger activity; however, a minor merger may be producing the spiral surface brightness enhancement in the center, which is a signature of sloshing and cold fronts. Also, no evidence for an X-ray counterpart of the central AGN is visible in the Chandra data, and no known AGN inside our FOV is bright enough to contribute significant hard emission relative to that of the thermal component. In addition to the radio jets, the core of the cluster is also host to an extended radio mini-halo (Murgia et al. 2009). As with radio halos and relics, IC emission may be detectable from the mini-halo if the magnetic field is small; Taylor et al. (1994) measured a lower limit of B  ≳  0.11–0.19 μG with Faraday rotation measure (RM) observations of the jet. The implied magnetic field strength, if we take as the IC flux that found with the 2–195 keV fit, is B ∼ 0.08 μG, roughly consistent with their field strength.

Zoom In Zoom Out Reset image size

But have we really detected IC from the cluster core? The significance of the non-thermal component completely disappears in the 3–195 keV fit; all three model combinations match the data equally well. Also, the 2T model formally provides a better fit to the 2–195 keV spectrum where the non-thermal component is detected. The second temperature component, ∼0.3 keV, is consistent with a low-temperature component of 0.11 keV observed by Clarke et al. (2004). Given these results, it is more likely that the non-thermal component is trying to mimic the low kT cool core component in the 2–3 keV range, since its significance disappears if this energy range is ignored. However, it is worth noting that the BAT data do generally support hard emission at higher energies, although at low S/N. Such hard emission could be due, on the other hand, to heavily obscured emission from an AGN within the FOV. The spatial distribution of BAT emission is consistent with that from a point source in all bands.

A1651 (Figure 5(b)). This cluster has a weak cool core, which means that while there is no significant temperature gradient in the center, the cooling time of the gas in the center is short (Hudson et al. 2010). Note that in a bimodal classification, A1651 would not be considered to have a cool core given its high central entropy of 90 keV cm² (Cavagnolo et al. 2009). Given the similarity between its BAT data and that of A2029, an obscured AGN of a similar flux could be responsible for the marginally detected positive flux in the higher energy bands. However, in this case the T+IC model is a significantly better fit than is the 2T model; Δχ² improves by 9 (2–195 keV) and 5 (3–195 keV) over the 1T and 2T models, respectively. If there were no hard excess, the probability that the six highest energy bands measure flux above the thermal component, given that BAT fluctuations are Gaussian, is (1/2)⁶, or 1.6%, which is not impressive in a sample of 59 clusters. The BAT spectrum is certainly suggestive, but considering that the excess is not significant at the 3σ level for the 3–195 keV fit, and only just at this level in the 2–195 keV fit—without including systematic uncertainties—we cannot claim to have detected a non-thermal component in this cluster. However, the evidence is perhaps strongest in this case, which is contrary to the expectation that such an excess is most likely in a merging cluster, particularly one with a radio halo, a relic, or a mini-halo, none of which are known to be associated with this cluster.

A2142 (Figure 5(c)). As the hottest cluster in the sample, the BAT is easily able to detect this cluster's high energy emission, which we might expect to exhibit a non-thermal excess since it also hosts a radio mini-halo (Giovannini & Feretti 2000). Both the T+IC and 2T models indicate that hard excess emission may be present; in the latter case, the second temperature component is unphysically high, acquiring the highest allowed temperature value. However, Nevalainen et al. (2004) estimate that two Seyfert galaxy nuclei within 17' of the cluster center contribute ∼30% of the hard-band emission detected by BeppoSAX; a similar amount of contamination would be expected in the BAT spectrum. SDSS J155829.36+271714.2 is bright in the XMM-Newton point-source catalog and likely contributes some portion of the flux measured in the BAT. Unfortunately, the XMM-Newton observation places this cluster right on the edge of the FOV, so over half (55.6%, based on a comparison with a pointed ROSAT PSPC image) of the soft-band emission is missing from the EPIC spectra. We rescale the XMM-Newton spectra to correct for the lost flux; the BAT source is equivalent to a point source, so it is not possible to correct the BAT emission for the XMM-Newton FOV. The correction to the XMM-Newton flux could be off by a sizable factor if the E > 2 keV emission is distributed differently than the E < 2 keV emission where ROSAT is sensitive. Chandra and Suzaku data of the core do exist, which may provide a measurement with less cross-calibration uncertainty. The significance of the non-thermal excess here is only at the 2σ level, mainly due to the poor statistics at XMM-Newton energies. While inconclusive, the BAT spectrum warrants further analysis using better data below 12 keV.

A3112 (Figure 5(d)). Using both Chandra and XMM-Newton data, Bonamente et al. (2007) have claimed to see both a hard and a soft excess that is consistent with a non-thermal origin. If this is the correct interpretation of these spectra, the IC excess would be clearly detectable in the BAT spectrum given our sensitivity. While a non-thermal component is detected in our joint fits, it has well below the predicted flux of Bonamente et al. (2007); our 3σ upper limit on the non-thermal normalization, using a photon index Γ = 1.8 that matches their best-fit value, is three times lower than their estimate. The quality of our 1T model fits is significantly less than for either the 2T or T+IC models; while those fits are of similar quality, the 2T fit yields physically reasonable temperatures and lower χ² values (Δχ² ∼ 3) than the T+IC model over both energy ranges. A non-thermal excess may in fact exist in this cluster, but a perhaps more likely scenario is that the ICM here is less isothermal than is typical in clusters, requiring several temperature components to adequately explain the cluster emission. The analysis of the Chandra data by Takizawa et al. (2003) in fact demonstrates the multi-temperature structure of this cluster, which may be exaggerated by significant gas cooling outside the core. In any case, the BAT data do not argue strongly in favor of an IC interpretation for the excess emission above ∼7 keV observed in the XMM-Newton data; as can be seen in Figure 5(d), the power-law component nearly ubiquitously overpredicts fluxes in the BAT spectrum. A more detailed exploration of the spatial and thermal structure at E < 12 keV is certainly warranted. This cluster is not known to host a diffuse radio halo, a relic, or a mini-halo.

A1367 (Figure 5(e)). This cluster hosts a radio relic in its outskirts (Gavazzi & Trinchieri 1983), and so IC emission is expected at some level in the radio relic region; however, the XMM-Newton/Swift extraction region does not contain the relic, so we are unable to address the strength of its magnetic field. Using RXTE, Henriksen & Mushotzky (2001) potentially detect a non-thermal component, although a two-temperature fit better describes their spectrum. The marginally detected IC emission we see is consistent with their non-thermal flux, whether we use a photon index of 2.0 or their value (based on the spectrum of the radio relic) of 2.9. Our 2T model fit, in the 2–195 keV band, is as good as the T+IC model fit, and given the marginally detected fluxes in the BAT bands, a 2T description of the ICM in this early-stage, forming cluster cannot be ruled out. However, the positive BAT fluxes and the consistency of our non-thermal fit with the analysis of the RXTE spectrum warrants future investigation of this cluster's hard X-ray emission.

A2589 & Fornax. Neither of the BAT spectra of these clusters, which also do not have a radio halo, a relic, or a mini-halo, show particular evidence that they have detected emission of any kind in any band. The first two bands of A2589's spectrum are just inconsistent with zero flux at the 1σ level, but a marginal detection in these bands is consistent with the thermal component. In both cases, the BAT spectrum is not sensitive enough to exclude the non-thermal component driven by the XMM-Newton data; since the BAT data do not further constrain the non-thermal component in these cases, we will not discuss these clusters further.

While some evidence for non-thermal emission is present in several of the HIFLUGCS clusters, in none of these cases is a significant excess indicated by both the BAT and EPIC spectra that could not plausibly be explained by a multi-temperature state of the ICM. In many cases, the BAT spectra simply lacked the S/N to meaningfully constrain the existence of excess emission; we therefore derive upper limits for a non-thermal component in our joint spectra. Three limits are presented for each energy range (2–195 keV and 3–195 keV) considered: a 90% confidence level limit including systematic uncertainties in f_CN and the EPIC backgrounds, as described in Section 2.1, and two 3σ limits, without systematic uncertainties included, for our fiducial photon index of Γ = 2 and for the best-fit value of Γ. After fitting for Γ, it is then fixed at that value when the upper limit is computed. The systematic terms are included in the 90% limits as described in Wik et al. (2009). Upper limits are reported as 20–80 keV fluxes in units of 10⁻¹² erg cm⁻² s⁻¹ in Table 5. Note that when Γ is much steeper than 2, the power-law component is constrained only by the low-energy spectrum and the 20–80 keV flux limits are not reflective of the sensitivity of the BAT survey. In some instances, usually for lower temperature clusters, the 90% limit exceeds the 3σ limits; in these fits, the systematic uncertainties in f_CN and/or the EPIC background dominate over the statistical uncertainty in the spectra. For example, in a low-temperature cluster lowering the EPIC backgrounds significantly hardens the spectra, while modifying f_CN such that already poorly constraining BAT fluxes are 10% higher will allow a much larger IC-like component to fit the data than would be allowed statistically. In hotter clusters, adjusting the background has less of an effect on their spectral shape, and because they are hot they tend to be more significantly detected by the BAT, so that modifying f_CN cannot drastically affect the non-thermal component.