THE DISSECTION OF ABELL 2744: A RICH CLUSTER GROWING THROUGH MAJOR AND MINOR MERGERS

Abell 2744 has been observed with Chandra using both the ACIS-I and ACIS-S chip arrays and we summarize the dates, exposure times, and Chandra ObsIDs in Table 1. The data were reprocessed starting with the level 1 event files and using CIAO version 3.4 with the latest gain and calibration files applied (CalDB version 3.5.5). All observations were performed in Very Faint (VF) data mode, allowing 5 × 5 pixel islands to be used in identifying cosmic ray events, significantly helping to reduce the particle background. Observation-specific bad pixel files were produced and applied and events flagged with ASCA grades 1, 5, and 7 were excluded. The data were filtered for periods of anomalously high background associated with flares. Standard procedures are followed in doing this, i.e., for the ACIS-I observations, point sources, and the region containing the diffuse cluster emission were excluded, and a light curve was extracted in the 0.3–10 keV range binned in 259 s intervals. The Sherpa lc_clean.sl routine was then used to identify for removal time intervals when the count rate deviated by more than 20% from the mean. No significant flares were detected, and the cleaned exposure times are listed in Table 1.

For background subtraction during spectral and imaging analyses, we use the blank sky observations ³ appropriate for the chip and epoch of the observation of interest. The background files were processed in the same manner as the observations, using the same background filtering, bad pixel, and gain files, and were reprojected to match the observations. The backgrounds were normalized so that the source and background counts in the 10–12 keV range matched. We checked for excess soft Galactic X-ray emission in the observations by extracting spectra from source-free regions and comparing to spectra extracted from the background files, finding no significant difference.

The left panel of Figure 1 shows a background-subtracted, 0.5–7 keV, exposure-corrected image of Abell 2744 which has been binned using the WVT binning algorithm of Diehl & Statler (2006), which is a generalization of the Cappellari & Copin (2003) Voronoi binning algorithm. The bin size is determined by the constraint that the signal-to-noise ratio (S/N) must be ∼5. Prior to binning, point sources detected with wavdetect were removed and the regions were filled via interpolation from a surrounding background region using dmfilth. The exposure map was generated using standard CIAO procedures ⁴ and accounts for the effects of vignetting, quantum efficiency (QE), QE non-uniformity, bad pixels, dithering, and effective area. The energy dependence of the effective area is accounted for by computing a weighted instrument map where the weights are computed from an absorbed MEKAL spectral model with Galactic absorption, temperature, redshift, and abundance set to the average cluster values obtained in Section 2.3.1. The right panel of Figure 1 shows the background-subtracted, exposure-corrected image of Abell 2744 with point sources included, and with a light Gaussian smooth (FWHM = 4'') applied. Abell 2744 is clearly a disturbed system, with a structure ∼750 kpc to the northwest of the X-ray surface brightness peak (labeled northwestern interloper). Within the main X-ray structure, there are a further three substructures: one ∼190 kpc south (labeled southern compact core), one ∼480 kpc to the north (labeled northern core), and another ∼220 kpc to the north of the X-ray surface brightness peak. The northern core furthest from the X-ray peak has a tail, which includes the second structure north of the X-ray peak, extending southward and curving westward toward the peak in the X-ray emission. The southern compact core has a blunted cone morphology with a prominent head and a fan-like tail of emission extending back toward the peak in the X-ray emission. There is also an edge in the surface brightness ∼250 kpc to the southeast of the southern substructure.

Zoom In Zoom Out Reset image size

The northwestern interloper appears asymmetric, with an edge just east of north and an extended tail toward the south. There also appears to be a bridge of emission connecting it to the main cluster emission. Based on a shorter Chandra observation, KD04 tentatively identified a cold front and bow shock on the eastern side of the northwestern interloper and also a fan-like feature extending toward the west. They concluded the northwestern interloper was moving from west to east on its way toward the cluster core. However, the deeper observations presented in Figure 1 do not confirm the existence of a bow shock and indicate the cold front on the eastern side is part of a larger northern edge, meaning the interpretation of KD04 may not be the correct one.

To emphasize the structures revealed in Figure 1 we present the residual significance map, which highlights low surface brightness structures and the unsharp masked image, which is useful for revealing sharp structures such as shocks and cold fronts, in Figure 2. To produce the significance map, we first fit the Chandra image with an azimuthally symmetric β model with best-fitting values of r_c = 471 ± 8 kpc, β = 0.96 ± 0.13, R.A. = 00^h14^m184, and decl. = −30°23'140. During the fitting of the β model, the northwest substructure was excluded, along with point sources. This β model accounts for the main cluster emission and is subtracted from Chandra image to produce a residual image which is then smoothed with a Gaussian kernel with σ = 3''. The resulting smoothed residual map is then divided by an error map to produce the residual significance map (see Neumann & Böhringer 1997; Owers et al. 2009c, for more details). The left panel in Figure 2 shows that the northwest interloper and its extended tail to the south are detected with high significance, while the northern core and southern compact core, along with the tail near the northern core, are also detected. The unsharp masked image is produced by dividing a background-subtracted, exposure-corrected image that was smoothed with a 44 Gaussian by the same image smoothed by 22'', corresponding to physical scales of 20 kpc and 100 kpc at the cluster redshift. The unsharp masked image is shown in the right panel of Figure 2, where it can be seen that the structure in the central cluster region is very similar to that seen in the residual significance map. However, the unsharp masked image accentuates the edge to the north of the northwestern interloper and also a finger of emission extending from the northwestern interloper toward the south.

Zoom In Zoom Out Reset image size

2.3.1. Temperature Maps

To better determine the nature of the features seen in Figures 1 and 2 and to identify other regions affected by the merger activity, we produce two maps of the projected temperature distribution for Abell 2744 using complementary techniques. The first temperature map is produced using a similar method to that developed by O'Sullivan et al. (2005) and Maughan et al. (2006), which is outlined in full in Randall et al. (2008). Briefly, the X-ray image is binned to a 1968 × 1968 pixel size and at the position of each pixel, source and background spectra are extracted in a circular region surrounding the pixel. The radius of this region is defined adaptively so that the extracted region contains 1500 background-subtracted counts. The process is expedited by extracting the response files from a coarser grid. For each spectrum, an absorbed MEKAL model is fitted in the energy range 0.6–9.5 keV, with the absorption column density set at the Galactic value, N_H = 1.61 × 10²⁰ cm⁻² (Dickey & Lockman 1990), the abundance set to the global value of Z_ave = 0.262 (see below), and the temperature free to vary. This temperature map is presented in the top left panel of Figure 3. We note that this method produces a temperature map with correlated pixels and the scale over which the pixels are correlated is not clear from the temperature map alone. Therefore, as a check we produced a second temperature map using the WVT algorithm to bin the image such that each (uncorrelated) spatial bin contains ∼1500 background-subtracted 0.5–7 keV counts. We select a number of regions of interest from the temperature map shown in the top left panel of Figure 3 (marked with crosses) and we require the WVT binning algorithm to place a bin centered on each of these regions. For each bin we extract observation-specific source spectra, corresponding count-weighted responses, and background spectra from the blank sky observations. The background spectra had their exposure times normalized so that the 9.5–12 KeV count rates match the observed rates. The spectra were re-binned so that there was at least 1 count per energy bin. The re-binned observation-specific spectra were fitted simultaneously over the 0.5–7 keV energy range within the XSPEC package (Arnaud 1996) using an absorbed MEKAL model (Kaastra 1992; Liedahl et al. 1995) with abundance, column density, and redshift fixed to those values used for the non-tessellated map. The Cash statistic was minimized during fitting. This tessellated temperature map is shown in the middle left panel of Figure 3. The tessellated and non-tessellated temperature maps are in excellent agreement.

Zoom In Zoom Out Reset image size

We have produced significance maps for both the tessellated and non-tessellated temperature maps which show how significant the spatial variations in temperature are when compared to the global average temperature, kT_ave. The significance, Δ(kT), is defined at each pixel as

$$\begin{equation} \Delta (kT) = \left \lbrace \begin{array}{@{}ll} \dsty \frac{(kT_{\rm pix} - kT_{\rm ave})}{{\sqrt{\sigma (kT_{\rm pix, hi})^2 + \sigma (kT_{\rm ave, lo})^2}}}, &kT_{\rm pix} < kT_{\rm ave}, \\ [13pt] \dsty \frac{ {(kT_{\rm pix} - kT_{\rm ave})}}{{\sqrt{\sigma (kT_{\rm pix, lo})^2 + \sigma (kT_{\rm ave, hi})^2}}}, &kT_{\rm pix} \ge kT_{\rm ave}, \end{array} \right. \end{equation} \tag{ 1 }$$

where kT_pix is the temperature map value for the pixel of interest, σ(kT_pix,hi) and σ(kT_pix,lo) are the upper and lower 68% confidence limits for kT_pix, respectively, and σ(kT_ave,hi)and σ(kT_ave,lo) are the upper and lower 68% confidence limits for kT_ave, respectively. The significance maps are shown in the top right and middle right panels of Figure 3 for the non-tessellated and tessellated maps, respectively. The global average temperature, kT_ave, was determined from fitting spectra extracted from a ∼900 kpc radius region containing the majority of the cluster emission with an absorbed MEKAL model, as outlined above for the tessellated maps, with the exception that the abundance was free to be fitted. We find kT_ave = 9.07 ± 0.15 keV, while the global average abundance Z_ave = 0.262 ± 0.025. These values are consistent with the values presented in Zhang et al. (2004).

The temperature maps reveal a rich diversity of patchy temperature structures, the most significant of which include the following. (1) A prominent cool ∼5 keV region associated with the northwest interloper where the coolest gas is coincident with the compressed X-ray isophotes, indicating there is a cold front there. We also note a hint of a cool swirl heading to the south of the northwestern interloper and coincident with the asymmetry seen in the X-ray brightness (Figure 2). (2) A cool ∼7.5 keV region associated with the southern compact core which is surrounded by hot >12 keV gas, the hottest of which resides just to the east of the substructure. (3) A ∼700 kpc ridge of hot >11 keV gas running toward the north which separates the northwestern interloper and northern core. (4) While the northern core and its tail do not appear to have a significantly different temperature to the average temperature, they do appear to be significantly cooler than the regions directly to the west and hotter than the regions directly to the east. This is especially clear when considering the tessellated map (middle left and right panels of Figure 3). (5) The peak in the X-ray surface brightness does not host the coolest gas, as would be expected in a relaxed, cool-core cluster.

2.3.2. Pseudoentropy, Pseudopressure, and Abundance Maps

From the non-tessellated temperature map, we have generated projected pseudopressure and pseudoentropy maps, similar to those presented in Finoguenov et al. (2005). These are shown in the bottom left and bottom right panels of Figure 3, respectively. The pseudoentropy at each pixel is K = kT_pixA^−1/3_pix, while the corresponding pseudopressure is P = kT_pixA^1/2_pix, where A_pix is the normalization determined during the fitting of the spectra corrected for exposure and normalized by the area of the extraction region. The northwestern interloper stands out as having the lowest pseudoentropy in the cluster complex, while the swirl of cool gas, hinted at in the temperature maps, is clearly seen as low entropy gas extending toward the south. The non-hydrostatic nature of Abell 2744 is further highlighted by the asymmetric, patchy pressure distribution which peaks at the X-ray surface brightness peak and has two high pressure fingers trailing off to the southeast and northeast.

Finally, we present an abundance map in the lower left panel of Figure 4. The technique outlined in Section 2.3.1 used to generate the non-tessellated temperature map was again applied. However, the abundance was free to vary during the fitting while, in order to allow more robust abundance determinations, the spectra were extracted from larger regions which contain ∼3000 counts. The abundances were measured relative to the solar photospheric values of Anders & Grevesse (1989). Included in the top left panel of Figure 4 is the corresponding temperature map which is in broad agreement with the temperature maps presented in the top and middle left panels of Figure 3. The most notable enhancements in abundance seen in the maps presented in the lower panels of Figure 4 are coincident with the southern compact core, the northern core, and the northwestern interloper. We identify the northern core as the highest metallicity region within Abell 2744. Furthermore, there is an anti-correlation between the abundance and the temperature insomuch as the hottest regions also appear to have the lowest abundances.

Zoom In Zoom Out Reset image size

We now explore the nature of several regions of interest which have been identified from Figures 1–3. The selected regions are overlaid onto the temperature and abundance maps in Figure 5. For each region of interest the source spectra, along with the corresponding backgrounds and responses, are extracted and fitted in XSPEC with an absorbed MEKAL model. During the fitting, the column density is fixed to the Galactic value while the temperature, abundance, normalization, and where the data are of a high enough quality, the redshifts are allowed to free to vary during the fit. The best-fitting parameters for each of the regions are presented in Table 2.

Zoom In Zoom Out Reset image size

2.4.1. Northern Core and Surrounding Structures

2.4.2. Southern Compact Core and Surrounding Structures

There are several regions of interest surrounding the southern compact core including the compact core itself (SCC1 in the left panel of Figure 5) and two hot regions identified in the temperature maps which straddle the core on the northeastern and southwestern sides (SCC2 and SCC3 in the left panel of Figure 5). While the detailed spectra extracted for the SCC2 region reveal the gas there is not significantly hotter than the global average temperature, the region to the northeast (SCC3) does contain significantly hotter gas which also has a very low abundance. We confirm that the southern compact core harbors gas which is both cooler and of a higher abundance than the global averages.

We attempt to quantify the strength of the surface brightness edge just southeast of the southern compact core (Figure 1) by fitting a spherically symmetric broken power-law density model centered at the center of curvature of the edge (using the same method as Owers et al. 2009b). The resulting fit gives a jump in density at the edge of 1.60^+0.15_−0.11. The surface brightness profile and the best-fitting broken power-law density model are shown in Figure 6. We note that in ObsIDs 7712 and 7915, the chip gap lies very close to the position of the edge. Therefore, we repeat the fit using only ObsIDs 8477 and 8557. We find that the best-fitting surface brightness model gives a density jump of 1.51^+0.17_−0.15, consistent with the measurement obtained using the full data set. The temperature measured just inside the edge (SCC4) is kT = 15.83^+3.80_−2.7 keV, while kT = 8.61^1.82_−1.42 keV in the SCC5 region just outside the edge. The change in pressure across this edge is (n_e,inkT_in)/(n_e,outkT_out) = 2.94^+0.98_−0.73, which indicates a significant pressure jump and thus the edge is likely a part of a shock front.

Zoom In Zoom Out Reset image size

2.4.3. Northwestern Interloper and Surrounding Structures

As noted in Section 2.2, the northwestern interloper contains a sharp surface brightness edge to the north. The temperatures just inside and just outside the front (regions NWI1 and NWI2 in the left panel of Figure 5) are 6.33^+0.83_−0.70 and 10.45^+2.52_−1.80 keV, respectively. Figure 7 shows the surface brightness profile across the edge, which we have fitted with a spherically symmetric broken power-law density model centered at the center of curvature of the edge (using the same method as Owers et al. 2009b) with a best-fitting density jump of 2.13^+0.20_−0.17 (uncertainties are 68% confidence limits). Providing the geometry assumed in the density model accurately represents the physical geometry of the edge, the temperature just inside the edge will be artificially boosted by hot gas from outside the edge which lies along the LOS. To account for this, we determine the deprojected temperature inside the edge by including a second MEKAL component in fitting the spectrum extracted from the NWI1 region. This second MEKAL component has temperature, normalizations, and abundance frozen to the values determined for the NWI2 region, with the normalizations multiplied by a constant that accounts for the different emission measures due to the different volumes probed (see Owers et al. 2009b for a detailed explanation of the procedure). The deprojected temperature in the NWI1 region is 5.09^+0.88_−0.71 and combining this with the measured density jump and the temperature measured in the NWI2 region, we find that the pressure across the front, given by (n_e,inkT_in)/(n_e,outkT_out) = 1.04^+0.32_−0.25, is continuous and, thus, the edge is a cold front.

Zoom In Zoom Out Reset image size

We extract spectra for four regions running from east to west across the northwestern interloper's low entropy tail (NWI3, NWI4, NWI5, and NWI6 in the left panel of Figure 5). The temperature map (top left panel in Figure 3) indicated that the low entropy tail may harbor cooler gas than the regions immediately to the east and west. Region NWI4 coincides with the lowest projected entropy, while region NWI5 coincides with the lowest temperature gas seen in the temperature maps in the vicinity of the low entropy tail. Regions NWI3 and NWI6 are the regions immediately east and west of the tail, respectively. The temperature and abundance measurements are listed in Table 2 and we find that the temperatures in NWI4 and NWI5 are indeed lower than those measured in NWI3 and NWI6, but only at the 68% confidence level.

2.4.4. Miscellaneous Regions with High Temperature and/or Low Abundance

Here, we outline the selection of galaxies targeted for spectroscopic observations, the observation and reduction of the spectra, the precision of our redshift measurements, and the spectroscopic completeness of our sample.

3.1.1. Photometric Catalog

3.1.2. AAOmega Observations

Observations were taken during the nights of 2006 September 12–18 using the AAOmega MOS on the Anglo-Australian Telescope (AAT; Saunders et al. 2004; Smith et al. 2004; Sharp et al. 2006). AAOmega consists of 392 2'' diameter fibers available for target allocation across a two-degree diameter field of view. During our observations, we used the medium-resolution (R ∼ 1300) 580V (blue arm) and 385R (red arm) gratings which provide a wavelength coverage of 3700– 8800 Å and spectral resolutions of 3.5 Å and 5.3 Å at 4800 Å and 7250 Å, respectively.

The target density greatly exceeds the maximum fiber packing density, requiring multiple configurations in order to obtain a high level of spectroscopic completeness, particularly in the dense core regions of the cluster. We utilize the simulated annealing algorithm (Miszalski et al. 2006) option within the AAOmega specific CONFIGURE ⁶ software to generate the fiber allocation files. This software operates most efficiently with an input catalog of ∼500–800 objects, particularly when configuring clustered fields. These requirements are met by partitioning the ∼1500 object primary target catalog into subsamples based on the target magnitude. The magnitude bins used were r_F < 19.63, 19.63 < r_F < 20.05, 20.05 < r_F < 20.37, 20.37 < r_F < 20.75, and 20.75 < r_F < 21.00.

The limiting magnitudes of the subsamples serve two purposes: (1) they allow a fairly random subsampling of the densest regions, thereby reducing the computation time for the CONFIGURE software, and (2) allow a shorter observation time to reach the desired S/N limit for the brighter subsamples, enabling a more efficient observing strategy. The limits are chosen to allow approximately two configurations per subsample, without undersampling fiber allocations. The configurations are done in sequence, starting with the brightest magnitude subsample and continuing to fainter subsamples. The unallocated objects in each subsample were propagated through to the next subsample in the configuration sequence. During the fiber configuration process, the CONFIGURE software attempts to allocate objects of higher priority ahead of lower priority objects, where prioritization is based on a user-defined rank assigned to each object. We required a high level of spectroscopic completeness in the central regions, so objects within 500 kpc of the cluster center were ranked highest. Rankings were decreased as a function of radius in 500 kpc radial bins out to a limit of 4 Mpc.

This method produced nine configurations and we summarize the observations of these configurations in Table 3. For each configuration, ∼40 fibers were allocated to blank sky regions for sky subtraction and a further eight fibers were allocated to bright stars for acquisition and guiding. Differential atmospheric refraction limits the length of contiguous exposures for any one configuration to about 3 hr; therefore, configurations requiring exposures longer than 3 hr were observed across several nights. For each configuration, we took a 5 s dome flat and 40 s FeAr arc lamp exposures for the purpose of flat fielding and wavelength calibration of the spectra. (Note that for configurations observed across several nights, night-specific calibrations were taken.) The data were reduced using the AAO 2dFDR pipeline software ⁷ which reduces the red and blue arm data separately, extracting sky-subtracted, flat-fielded and wavelength-calibrated data for each exposure frame. The reduced frames are then co-added (using a weight derived from the total flux in each frame), at which point cosmic rays are identified and rejected. After co-addition, the red and blue-arm spectra are spliced together. At this point, the data were processed using an IDL code which reduces the residuals left from sky subtraction via a Principal Component Analysis (PCA) process which is adapted from software developed by Wild & Hewett (2005) and E. Wisnioski (2008, private communication). See Drinkwater et al. (2010) for a more detailed explanation.

Table 3. Summary of the AAOmega Observations

Field	Date	rF Limit	Texp	Nobjects	NRedshift	Seeing
1	2006 Sep 12	19.63	5 × 1800 s	192	180	12
2	2006 Sep 13, 14	19.63	4 × 1800 + 3 × 1800 s	134	126	14
3	2006 Sep 12	20.05	10 × 1800 s	172	161	1.2–14
4	2006 Sep 13, 14, 15	20.37	4 × 1800 + 6 × 1800 + 5 × 1800 s	181	147	1.0–15
5	2006 Sep 13, 14, 15	20.37	1 × 1800 + 5 × 1800 + 4 × 1800 s	130	114	1.0–15
6	2006 Sep 15, 16	20.75	5 × 1800 + 7 × 1800 s	190	152	1.0–15
7	2006 Sep 16, 17	20.75	5 × 1800 + 5 × 1800 s	133	109	10
8	2006 Sep 17, 18	21.00	(10 × 1800 + 1 × 1200) + (6 × 1200 + 1 × 800) s	191	157	10
9	2006 Sep 18	21.00	(12 × 1800) s	149	113	10

Download table as:  ASCIITypeset image

The reduced, combined, PCA-corrected data are then assigned redshifts using the RUNZ code, which was written by Will Sutherland for the 2dF Galaxy Redshift Survey (2dFGRS; Colless et al. 2001). The RUNZ code has been adapted for use with AAOmega spectra and significantly enhanced from its earlier versions. In particular, emission-line-redshift measurements have been significantly improved for the WiggleZ survey (Drinkwater et al. 2010). The code measures redshifts using two methods: (1) for absorption line spectra the cross-correlation method of Tonry & Davis (1979) is employed with a suitable library of template spectra and (2) for emission-line galaxies the redshifts are measured for each line by fitting a Gaussian, with the final emission-line redshift being determined from the variance-weighted mean of the redshifts determined for each emission line. The quality of the assigned redshift was determined by visual inspection and an integer value classification, Q, was assigned, where Q ranges from 1 to 6, using the same scheme adopted for the 2dFGRS, as described in Owers et al. (2009a, 2009c). Briefly, objects with Q ⩽ 2 are assigned no redshift or an unreliable redshift, objects with Q = 3, 4, or 5 have reliable redshifts (with Q = 5 having a template-like quality spectrum) and objects with Q = 6 are stars or non-extragalactic objects. The AAOmega observations produced reliable (Q = 3, 4, or 5) redshifts for 983 extragalactic objects and 276 stars.

3.1.3. Redshift Uncertainties

Of the 983 Q = 3, 4, or 5 AAOmega observations, there were 38 galaxy spectra with repeat observations. These repeat observations were used to check the uncertainties assigned by RUNZ to redshift measurements. We use a robust biweight estimator to determine the mean and dispersion of the distribution of redshift differences, where the redshift measurement with the lower quality (defined by ether its lower Q value, or higher error measurement) is subtracted from the higher quality measurement. We find a mean difference of $\overline{\Delta cz}=1\pm 13$ km s^-1 and a dispersion of 87 ± 21 km s^-1 from which we can infer a single redshift measurement uncertainty of 61 ± 15 km s^-1. The median value of the individual redshift uncertainties assigned by RUNZ to non-stellar objects was 66 km s^-1 (after excluding six outliers with error values greater than 1000 km s^-1), which is consistent with the uncertainty derived from the repeat observations.

Independent checks on the redshifts and their uncertainties can be made by cross-correlating the AAOmega catalog with the Couch & Sharples (1987), Couch et al. (1998), Boschin et al. (2006), and Braglia et al. (2009) redshift catalogs. Further to these catalogs, we add archival 2dF data, which is reduced using 2dFDR and redshifted using the RVSAO package in IRAF (Kurtz & Mink 1998). The results of the comparisons are presented in Table 4 where the columns are the following: origin of external redshift catalog for comparison (Column 1), number of redshifts in common with the AAOmega catalog, N_match, where catalog members are deemed to be matches if their positions differ by less than 3'' (Column 2), the mean redshift difference, $\overline{\Delta cz}=\overline{cz_{{\rm ours}}-cz_{{\rm ext.}}}$, as determined by the biweight estimator (Column 3), the dispersion in the Δcz distribution, σ(Δcz), determined with the biweight estimator (Column 4), the redshift uncertainty for the external redshift measurements (Column 5), the quadrature difference between σ(Δcz) and the external redshift error which gives an external estimate of the uncertainty on the AAOmega measurements (Column 6), and the ratio of the external uncertainty measurement to the internal uncertainty measurement derived above (Column 7). We note that the results in Table 4 show that our internal redshift uncertainties are systematically underestimated by a factor of ∼1.8. We conclude that the redshift precision of the AAOmega measurements are ∼110 km s^-1.

Table 4. Comparison of AAOmega Redshift Measurements with Literature Values

Literature Source	Nmatch	$\overline{\Delta cz}$	σ(Δcz)	Literature Redshift	Quadrature Difference	Ratio
		( km s-1)		Uncertainty ( km s-1)	( km s-1)	( km s-1)
Braglia et al. (2009)	60	88+39−38	289+46−40	276	86 ± 156	1.4 ± 2.6
Boschin et al. (2006)	56	−18+23−24	172+19−18	90	147 ± 22	2.4 ± 0.7
Couch et al. (1998)	29	177+24−24	125+23−18	90	87 ± 32	1.4 ± 0.6
Couch & Sharples (1987)	28	−89+27−27	151+24−26	100	113 ± 32	1.5 ± 0.7
2df 2003	24	−53+27−26	136+20−23	50	127 ± 21	2.1 ± 0.6

Note. The errors presented for $\overline{\Delta cz}$ and σ(Δcz) are determined from the 68% confidence limits based on 10,000 bootstrap resamplings of the data.

Download table as:  ASCIITypeset image

3.1.4. Combined Redshift Catalog

3.1.5. Spectroscopic Completeness

In the forthcoming sections, we will use the redshift catalog to search for substructure in Abell 2744. Therefore, in order to be confident in the results, it is important that the spectroscopic completeness is well understood, particularly for the interpretation of maps of the projected density of the member galaxies. We use the combined redshift catalog to determine the spectroscopic completeness, which is defined as the ratio of the number of objects with reliable redshift determinations to the number of objects in the parent photometric catalog. In Figure 8, we present the spectroscopic completeness as a function of radius, r_F magnitude and a combination of radius and r_F magnitude. The plots show that we have obtained a high level of completeness at all cluster-centric radii and for all magnitude ranges. However, the spectroscopic completeness is systematically lower within the central ∼1 Mpc (although we note it is still excellent and remains above ∼70% for all magnitude ranges). This is due to a combination of fiber separation limits (∼30'') and the high galaxy surface density in the central regions, which means a small number of objects could not be observed.

Zoom In Zoom Out Reset image size

Figure 9 shows two images of the spatial distribution of the spectroscopic completeness for the magnitude limits r_F < 20.5 (left panel) and r_F < 21 (right panel). The images were generated by first producing an adaptively binned image of the spatial distribution of the galaxies in the photometric catalog where each spatial bin contains ∼10 galaxies. We made use of the WVT binning algorithm by Diehl & Statler (2006), which is a generalization of Cappellari & Copin's (2003) Voronoi binning algorithm. The same spatial binning was then applied to the galaxies in the spectroscopic catalog to produce a second image. The ratio of the binned spectroscopic to photometric images gives the spectroscopic completeness map in Figure 9. The maps reveal the high level of spectroscopic completeness we have obtained with our observations and, in particular, for r_F < 20.5 we obtain ≳90% completeness over the majority of the 6 Mpc diameter field. For r_F < 21 we obtain ≳80% completeness over the majority of the 6 Mpc diameter field. The systematically lower spectral completeness within 1 Mpc noted in the radial profile shown in the right panel of Figure 8 is due to a region of low completeness confined to the northwest of the center. This patch of lower spectral completeness needs to be kept in mind in the forthcoming sections.

Zoom In Zoom Out Reset image size

3.1.6. Cluster Membership

Robustly distinguishing cluster members from foreground and background interlopers is a critical first step when conducting a dynamical analysis. Membership allocation is a two-step process where the initial step involves crudely separating cluster members and projected interlopers in redshift space only. This is achieved by selecting galaxies within the redshift window of cz_clus ± 10, 000 km s^-1 (where z_clus = 0.306; Boschin et al. 2006) which, as the left panel in Figure 10 shows, defines a peak in redshift space associated with Abell 2744. The second part of the process refines the membership allocation by using a slightly modified version of the "shifting gapper" first employed by Fadda et al. (1996). This method utilizes both radial and peculiar velocity information to separate interlopers from members as a function of cluster-centric radius (defined with respect to the bright cluster galaxy (BCG) closest to the X-ray peak at R.A. = 00^h14^m20738, decl. = −30°23'5990), and is described in detail in Owers et al. (2009c). Briefly, the data are binned radially such that each bin contains at least 40 objects. Within each bin, galaxies are sorted by their peculiar velocity and interlopers are rejected based on the size of the gap in velocity between adjacent (in peculiar velocity) objects. The limit of 40 objects per bin is somewhat subjective and is chosen to minimize both the radial annulus used and the likelihood of the algorithm rejecting either high-velocity bona fide cluster members or coherent substructures associated with merger activity. Binning by 40 objects produces the best results, as shown in the right panel of Figure 10 where we present the results of the shifting gapper membership allocation procedure. Figure 10 shows that within a cluster-centric radius of 3 Mpc the cluster members are well separated from the surrounding interlopers. Beyond 3 Mpc the separation is less clear, so we exclude the galaxies at these larger radii during the analysis. After culling the interlopers, 343 bona fide cluster members remain from which we determine a cluster redshift of z_clus = 0.3064 ± 0.0004 and a velocity dispersion of σ_clus = 1497 ± 57 km s^-1 using biweight estimators of location and scale (Beers et al. 1990), where the quoted errors are 1σ values determined using the jackknife resampling technique.

Zoom In Zoom Out Reset image size

Also plotted in the right panel of Figure 10 are the means (filled circles) and 3σ limits (filled diamonds) for each radial bin, determined using biweight estimators, showing that all excluded interlopers have peculiar velocities exceeding the 3σ limit at the radial bin of interest. The inadequacy of a simple 3σ clipping using only velocity information is highlighted in the right panel of Figure 10, where the dashed lines show that the 3σ_clus limits fail to reject several interlopers at larger cluster-centric radii.

Cluster mergers can induce significant distortions in the observed velocity distribution which, when a cluster is dynamically relaxed, is well approximated by a Gaussian. Therefore, searching for evidence of substructure using velocity information is generally accomplished by testing for departures from a Gaussian shape. Here, we use the method first outlined by Zabludoff et al. (1993) which approximates the velocity distribution by the summation of three Gauss–Hermite functions. The method is described in detail in Owers et al. (2009c). The coefficient of the first term, h₀ ≈ 1, multiplies the best-fitting Gaussian contribution to the distribution while the second and third terms, with coefficients h₃ and h₄, approximate the third- and fourth-order asymmetric and symmetric deviations from a Gaussian shape, similar to skewness and kurtosis measurements but less sensitive to outliers in the distribution. For Abell 2744, we find the coefficients have values of h₃ = 0.07 and h₄ = 0.06. The significance of these values is determined by generating 10,000 Monte Carlo realizations of Gaussians with N = 343 data points having mean and standard deviation equal to the best-fitting value derived from the data. We find that values of |h₃|>0.07 occur in only 6% of the realizations, while values of |h₄|>0.06 occur 8% of the time. Thus, there appear to be mildly significant symmetric and asymmetric deviations from Gaussianity present in Abell 2744's velocity distribution.

The positive value for the h₄ term is likely to be due to the contribution of galaxies in the cluster outskirts, which follow more radial orbits and hence produce a more peaked velocity distribution (Merritt 1987), rather than the presence of substructure. Because we are interested in detecting structure related to a cluster merger and not due to orbital anisotropy, we re-determine the Gauss–Hermite coefficients using only data from the central 1.5 Mpc region. There are 238 galaxies within this region in our sample and for these we measure h₃ = 0.12 and h₄ = −0.03. The observed h₃ coefficient is highly significant with |h₃|>0.12 occurring in less than 1% of the realizations, while the h₄ term is not significant and |h₄|>0.03 occurs in 47% of the realizations, verifying that the mildly significant positive h₄ value measured above is indeed due to the contribution of the galaxies at larger cluster-centric radii.

As noted previously, the velocity distribution in Abell 2744 is bimodal (Girardi & Mezzetti 2001; Boschin et al. 2006; Braglia et al. 2007). This bimodality causes the significantly positive h₃ term and Figure 11 clearly shows a secondary substructure with a peak at v_pec ∼ 2000 km s^-1. To further quantify the significance of this bimodality, we use the Kayes Mixture Model (KMM) algorithm (Ashman et al. 1994) to fit two Gaussians to the velocity distribution of the members residing within a cluster-centric radius of 1.5 Mpc. As input for the main component, we set a mean μ_1,init = 0, dispersion σ_1,init = 1300 km s^-1, and estimate the fractional contribution of the number of galaxies belonging to the main component with respect to the entire distribution as f_1,init = 0.7. Likewise, for the secondary component we set μ_2,init = 2000 km s^-1, σ_2,init = 500 km s^-1, and f_2,init = 0.3. As output, the algorithm returns μ_1,out = −596 km s^-1, σ_1,out = 1261 km s^-1, and f_1,out = 0.79 for the main component and μ_2,out = 2316 km s^-1, σ_2,out = 581 km s^-1, and f_2,out = 0.21 for the secondary component. It is noted that the outputs generated by the KMM algorithm are robust to changes in the initial input parameters. The resulting Gaussians are overplotted on the observed velocity distribution in the right panel of Figure 11. Since we are using the algorithm for a heteroscedastic (i.e., where the variances differ) case, the estimate of the significance of the likelihood ratio test statistic (LRTS) comparing a bimodal fit to the null-hypothesis unimodal fit given by the algorithm is unreliable (Ashman et al. 1994). To reliably determine the significance of the LRTS, we produce 10,000 randomly sampled Gaussian distributions with 238 data points, mean −38 km s^-1, and dispersion 1632 km s^-1 (i.e., mean and standard deviation which best describe the observed velocity distribution). For each random data set, two Gaussians are fitted using the KMM algorithm, as was done for the observations, where the initial inputs are set to the best-fitting outputs for the observed velocity distribution listed above and the LRTS is computed. Comparison of the observed LRTS to the distribution of 10,000 LRTS produced by the random unimodal Gaussian data sets allows the significance of the observed LRTS to be reliably determined. The observed LRTS is larger than 99.99% of the LRTSs produced in the 10,000 realizations and we conclude that a bimodal fit is significantly better than a unimodal one.

Zoom In Zoom Out Reset image size

Having shown that the velocity distribution in Abell 2744 is bimodal, we now search for further evidence of merger-related substructure using two-dimensional spatial information. To that end, we have used the positions of the 343 spectroscopically confirmed members to produce maps of the smoothed galaxy surface density (Figure 12). We used a variable width Gaussian smoothing, where the width, σ, at each pixel is the projected distance to the tenth nearest galaxy, limited to a minimum of 90 kpc and a maximum of 500 kpc in high- and low-density regions, respectively. As detected previously (Boschin et al. 2006; Braglia et al. 2009), two distinct substructures inhabit the central regions of the cluster—the most significant one is centered close the cluster center (which was arbitrarily defined in Section 3.1.6 to be the position of the BCG nearest the peak in the X-ray surface brightness) and a second substructure ∼600 kpc to the north. At larger scales, the distribution of galaxies appears smooth, apart from a structure ∼2.5 Mpc to the west and a mild asymmetry to the northwest.

Zoom In Zoom Out Reset image size

Given the bimodality detected in the velocity distribution, we search for any correlation between the two velocity peaks and the two substructures seen in the two-dimensional galaxy density map for the spectroscopically confirmed members (top panels of Figure 12). To achieve this, we plot separately the spatial distributions of the galaxies allocated to the two KMM velocity partitions in Section 4.1 and smooth the resulting distributions with an adaptive circular top hat filter with radius set such that the smoothing region contains 10 galaxies. These plots are shown in Figure 13 where the left and right panels show the distributions of galaxies assigned to the lower and higher velocity partition, respectively. The spatial distribution of the galaxies assigned to the dominant (both in terms of number of galaxies and velocity dispersion) lower velocity partition has two local peaks in galaxy surface density. The most significant of these two peaks is located coincident with the northern substructure seen in Figure 12, while the less significant one is approximately coincident with the more central density peak seen in Figure 12. The spatial distribution of the high-velocity partition also has its peak density approximately coincident with the central density peak seen in Figure 12. This indicates that the density peak seen close to the center in Figure 12 is not the center of the cluster, but is caused by the alignment of two substructures separated by a large velocity and projected along our LOS.

Zoom In Zoom Out Reset image size

Inspecting the completeness map Figure 9 reveals that galaxies in the region of the northern peak are underrepresented in our spectroscopic sample. Presumably, improving the sample completeness in this region would enhance the local density peak. Thus, the northern peak is likely to be even more significant than it appears to be in Figure 13.

The top right panel in Figure 12 shows a close up of the smoothed member galaxy surface density in the central 1 × 1 Mpc region with Chandra X-ray contours overplotted. Neither of the local peaks in galaxy surface density coincides with a peak in X-ray surface brightness. Of particular concern is the lack of any significant overdensity coincident with the northwestern interloper where Boschin et al. (2006) and Braglia et al. (2009) both detected substructure using photometrically selected cluster members. As noted in Section 3.1.5, the region just northwest of the center has the lowest spectroscopic completeness and this may be a contributing factor in the non-detection of substructure in this region. To test this, we obtain the photometric catalog ⁸ of Busarello et al. (2002), which covers the central 2.3 × 2.3 Mpc region, and utilize their publicly available photometric redshifts to define cluster membership as per the definition in Busarello et al. (2002). The lower left panel of Figure 12 shows the adaptively smoothed galaxy surface density for the photometric members with R ⩽ 21 (where 70 ⩽ σ ⩽ 400 kpc). At this magnitude limit, which approximately matches our spectroscopic limit of r_F = 21, there is qualitative agreement with the structures detected in the galaxy surface density map for the spectroscopically confirmed members. Therefore, the non-detection of a significant local overdensity coincident with the northwestern interloper, as detected by Boschin et al. (2006) and Braglia et al. (2009), is not due to the lower spectroscopic completeness, but rather is due to the fact that the majority of the galaxies contributing to the overdensity there are fainter than our limiting magnitude. This can be seen in the lower right panel of Figure 12 where we have plotted the adaptively smoothed galaxy surface density for the photometric members down to the completeness limit of R = 22.3 (where 50 kpc ⩽ σ ⩽ 400 kpc). This map reveals the presence of an overdensity approximately coincident with the northwestern interloper, consistent with the maps of Boschin et al. (2006) and Braglia et al. (2009). This result is consistent with Andreon (2001) who, using deep K-band imagery, find significantly more dwarf galaxies and significantly fewer bright galaxies in this northwestern region when compared to the central cluster region (which is roughly coincident with our defined cluster center). Confirmation that this overdensity is part of the Abell 2744 system requires deeper spectroscopic observations.

We now combine the spatial and velocity information to determine if there is any correlation between the substructures detected in the two-dimensional spatial and one-dimensional velocity distributions above. This is achieved by making use of the κ-test which was successfully employed by Colless & Dunn (1996) to detect substructure in the Coma cluster. The κ-test searches for local departures from the global velocity distribution around each cluster member galaxy. This is accomplished by using the Kolmogorov–Smirnov (KS) test which determines the likelihood that the velocity distribution of the $n=\sqrt{N}$ (where N is the number of cluster members) nearest neighbors around the galaxy of interest and the global cluster velocity distribution, which includes the remaining N − n galaxies, are drawn from the same parent distribution. The likelihood is determined by measuring the maximum separation of the two cumulative distribution functions, D_obs and determining the probability that the D statistic is larger than D_obs for the given sample sizes, P_KS(D > D_obs). The global measure of the dynamical substructure present within the cluster, κ, is determined by summing the negative log likelihood for each galaxy over the entire sample, i.e.,

$$\begin{equation} \kappa = \sum -{\rm log}P_{\rm {KS}}(D > D_{{\rm obs}}). \end{equation} \tag{ 2 }$$

The significance of κ is determined by performing 10,000 Monte Carlo realizations where, for each realization, the observed positions of the galaxies are retained, while the galaxy velocities are shuffled randomly, removing any correlation between positions and velocities. The observed κ value is then compared to the distribution of the 10,000 Monte Carlo realizations of κ. A value as high as the observed value of κ = 481 does not occur in the 10,000 realizations, which have a distribution that is well modeled by a log-normal distribution with μ(ln κ) = 4.95 and σ(ln κ) = 0.19. Thus, the observed κ value lies ∼6.6σ from the mean of the realizations and the upper limit on the probability of observing a value this high by chance is 10⁻⁴.

Figure 14 shows the results of the κ-test in the form of a "bubble plot." At the position of each member galaxy, a circle is plotted with radius ∝−log  P_KS(D > D_obs). Clusters of large bubbles reveal regions where the local dynamics are different from the global cluster dynamics; we define significantly large bubbles as those with radii that occur in less than 1% of the 10,000 realizations and these are emboldened in Figure 14. The bubbles are also color-coded based on the sign of the galaxy's peculiar velocity, where red and blue mark positive and negative v_pec, respectively. Overplotted are the galaxy surface density contours taken from the adaptively smoothed image shown in the top left panel of Figure 12. There are two clear dynamical substructures detected in the central ∼1 Mpc; one coincident with the cluster center and another to the north, coincident with the second projected galaxy overdensity. At larger radii, we also detect the two substructures to the northwest and south first noted by Braglia et al. (2007).

Zoom In Zoom Out Reset image size

Radial gradients in the cluster velocity dispersion are expected and, in particular, the velocity dispersion is expected to decrease at larger radii due to a combination of the radial orbits of infalling galaxies and the decreasing escape velocity of the cluster. This can cause spurious results in three-dimensional tests, such as the κ-test used here, which are not related to dynamical substructures (Pinkney et al. 1996). For this reason, we re-ran the κ-test considering only those galaxies residing within a central 1.5 Mpc radius region. Again, we find the observed value of κ = 200 does not occur in any of the 10,000 Monte Carlo realizations and lies ∼3.6σ from the mean of the simulated κ distribution, which has μ(ln κ) = 4.58 and σ(ln κ) = 0.20.

The κ-test is an excellent tool for locating dynamically distinct substructure; however, it provides no insight into the origin of the differences in the kinematics of the galaxies within the substructures. As a first step toward visualizing and characterizing the substructure revealed by the κ-test in Figure 14, we present Figure 15 which shows the LOS velocity field. The value of each pixel in the LOS velocity field map is the trimmed mean of the velocity distribution for the 15 nearest neighbors in projection. Only pixels where the radius to the 15th nearest neighbor is ⩽1000 kpc are presented in Figure 15. The trimmed mean is measured by first determining the median and median absolute deviation (which are insensitive to outliers) as proxies for the mean and standard deviation for the distribution, trimming those points that are further than 3σ from the median and calculating the mean of the remaining points. The velocity field is complex and, most significantly, there is clear evidence for velocity structure within the central 1 Mpc region where a low-velocity (∼−1300 km s^-1) structure is seen to the north, while a high-velocity structure (∼2400 km s^-1) is seen to the south. These two structures coincide with the two substructures revealed by the κ-test in Figure 14.

Zoom In Zoom Out Reset image size

We now attempt to disentangle these substructures and ascertain their dynamics. To do this, we again utilize the KMM algorithm of Ashman et al. (1994) as in Section 4.1; however, here we use both spatial and velocity information to partition the cluster into its dynamical substructures. Since we are mainly interested in the dynamics of the central region, where we have Chandra data, and the KMM algorithm can be biased by outliers, we consider only members within a cluster-centric radius of 1.5 Mpc for the KMM analysis. As input, the KMM algorithm requires the number of partitions needed to describe the data, as well as the approximate mixing proportions and estimates of the means and covariance matrices for each partition. One of the major uncertainties when using this method is in determining the number of partitions required to adequately describe the data. We implement two strategies to overcome this uncertainty. First, we use the results from the κ-test, in combination with the galaxy surface density maps, to guide us in selecting the number of likely substructures. Second, we utilize the hypothesis testing capabilities of the KMM algorithm by measuring the LRTS, which gives a measure of the improvement in the fit for a complex, g-partition, model over a simpler, g₀-partition, model. In Ashman et al. (1994), the simpler model was a unimodal one (i.e., g₀ = 1) where the significance of the LRTS, expressed as a P-value, was estimated by comparison with a χ² distribution, which is only suitable when comparing two homoscedastic (i.e., the same variance), univariate models—clearly not the case here. Furthermore, we would like to assess how significantly the fit is improved by the addition of one partition, i.e., the g-partition case over the g₀ = g − 1 partition case, in order to decide how many partitions to use. Therefore, we employ a parametric bootstrap resampling technique to determine the P-values as follows. We use KMM to fit the g₀- and g-partition three-dimensional Gaussian models to the data. We then produce 5000 bootstrapped samples by randomly sampling the best-fitting g₀-partition three-dimensional Gaussian mixture model. To each of these bootstrap samples, we re-fit the g₀- and g-partition models, using the means and covariances of the models fitted to the data as initial input estimates for the KMM algorithm and remeasure the LRTS. The observed value for the LRTS is then compared to the null distribution of bootstrapped LRTS values to determine the P-value.

Figure 14 reveals two clear substructures: one with v_pec ∼ 2300 km s^-1 close to the cluster center, which we label Clump A and one with v_pec ∼ −1600  km s^-1 and ∼700 kpc north of the cluster center, which we label Clump B. There is also a possible substructure approximately spatially coincident with Clump A, but with a v_pec similar to that of Clump B, which we call Clump C. These partitions were used as input for the KMM algorithm. Thus, we fit three models: a g = 2, a g = 3, and a g = 4 partition model and at each stage we determine the P-value for the null hypothesis g₀ = g − 1 fit, as outlined above. To determine the initial estimates for input into the KMM algorithm, we take advantage of the fact that each of these clumps houses a bright (r_F < 17.8) galaxy which has v_pec at the peak of the velocity distribution (Figure 14). We assume that these bright galaxies lie at the spatial and dynamical centers of each partition and crudely define partition membership as any galaxy within v_pec = ±1500  km s^-1 and radius ⩽350 kpc of the central galaxy. The means and covariance arrays for these initial partitions serve as the initial estimates for the KMM algorithm. The results of the g = 2, 3,  and 4 partition KMM fits are presented in Figure 16. The P-values for these fits are P_g=2 < 0.0002, P_g=3 = 0.02,  and P_g=4 = 0.93. Thus, the g = 3 partition fit, with Clumps B and C combined, is the statistically preferred one, despite a spatial separation of ∼700 kpc between the two bright galaxies in these clumps.

Zoom In Zoom Out Reset image size

In their analysis of the dynamics of Abell 2744, Boschin et al. (2006) detect the northern major remnant core in both the two-dimensional and three-dimensional analyses but discard the hypothesis that it is the undisturbed remnant core of the main cluster. The combination of the new Chandra X-ray data and comprehensive AAOmega spectroscopy reveals that the substructure north of the central regions of Abell 2744 is more significant than previously thought. The strongest evidence for this assertion comes from Figure 13 which shows that when the contribution to the galaxy surface density from the high-velocity southern component is removed, the most significant peak in the galaxy surface density is coincident with the northern substructure. In Section 4.3, we show that this galaxy surface density enhancement is detected as a dynamically distinct substructure with v_pec ∼ −1600 km s^-1 and σ_v ∼ 800 km s^-1. Furthermore, the deeper Chandra observations resolve a gas substructure and reveal a tail of gas trailing to the south which curves toward the west (Figure 1). The abundance map presented in Section 2.3 (Figure 4) shows that this gas substructure has significantly higher metallicity than the global cluster value, while the analysis of detailed spectra (NC1 in Table 2) show that this gas structure has a significantly lower temperature (kT ∼ 7 keV) than the global cluster temperature.

Observations of relaxed clusters show that the central core regions are characterized by a steep decline in gas temperature and a corresponding increase in metallicity (Vikhlinin et al. 2005). Given these observations and the thermodynamic properties outlined above, we interpret the northern gas substructure as the remnant of a cool core which has survived the merging activity. In addition to this, the large velocity dispersion, the significant number of galaxies in this region, along with the fact that this structure harbors one of the bright cluster members (with v_pec ∼ −1200 km s^-1), we postulate that this northern component has a significant mass and is the core of the more massive system involved in the central merger in Abell 2744.

This structure, located coincident in projection with the high-velocity southern substructure (Section 5.3), but separated in peculiar velocity by ∼4000 km s^-1 as discussed in Section 4.3 (Figure 14 inset), has previously been interpreted as being the core of the main negative v_pec component in the Abell 2744 system (Kempner & David 2004; Boschin et al. 2006). This interpretation is supported by the local peak in the surface density of the galaxies and the presence of a luminous galaxy with v_pec ∼ −1600 km s^-1. However, as discussed in Section 4.2 (see also Figure 13), the peak in the projected galaxy density here is artificially enhanced by the high-velocity southern substructure. When this is accounted for, it can be seen that the more significant peak in the galaxy surface density of the negative v_pec component is in fact the northern substructure, which we interpret as the remnant core of the main cluster, as discussed in Section 5.1.

Despite not being the core of the negative v_pec component, this structure is a dynamically distinct component with a v_pec ∼ −1600 km s^-1. In Section 4.3, we found that adding a component representing this structure to the KMM analysis failed to improve the fit significantly over the simpler three-partition fit. Therefore, we tentatively identify this structure as tidal debris stripped from the main cluster (the remnant core of which now lies mainly to the north) during the core passage phase of the merger with the high-velocity southern substructure.

The southern compact core stands out as a prominent surface brightness enhancement in the Chandra images (Figures 1 and 2). It is conspicuous in the thermodynamic maps presented in Section 2.3 as a region with significantly cooler, higher metallicity, and lower entropy gas (Figures 3 and 4). This provides strong evidence for the southern compact core being the remnant cool core of a merging substructure. In Section 4.1, we found that the cluster peculiar velocity distribution was bimodal, containing a high-velocity component. In Section 4.2, we explored the spatial distribution of this high-velocity component, finding that it is spatially compact, while the combination of spatial and velocity information in Section 4.3 confirmed the existence of this structure as a dynamically distinct, spatially compact entity which we label as the southern minor remnant core in Figure 17. The proximity of the southern minor remnant core to the southern compact core indicates that they are associated; the bright galaxy accompanying the southern minor remnant core is located only ∼140 kpc to the southeast of the X-ray peak due to the southern compact core. The peculiar velocity of the southern minor remnant core, obtained with the KMM analyses in Sections 4.1 (for the one-dimensional velocity analysis) and 4.3 (for a full three-dimensional velocity plus spatial information analysis), is v_pec = 2300–2500 km s⁻¹. We find evidence that the southern compact core also has a high peculiar velocity by considering the shock front to the southeast (Section 2.4.2). The Rankine–Hugoniot shock jump conditions can be used to derive a Mach number for the shock from the density jump and thus a shock velocity (e.g., Equation (14) of Markevitch & Vikhlinin 2007; Landau & Lifshitz 1959). For the measured density jump of 1.60^+0.17_−0.11, we determine a Mach number of M = 1.41^+0.113_−0.08. We can check this result by using Equations (4) and (14) of Markevitch & Vikhlinin (2007) and the temperature measurements taken on either side of the front to obtain M = 1.81^+0.49_−0.37, consistent within the errors. The sound speed for a ∼8.6 keV gas is ∼1523 km s⁻¹; therefore, the shock velocity is ∼2150 km s⁻¹, using the Mach number derived from the density jump.

We note that this method does not give the total velocity of the shock, rather it gives the shock velocity in the plane of the sky. This is due to the following. In order that the shock front be observable as an edge in the surface brightness, our LOS must be tangent to some point on the shock front surface. Assuming we know the radius of curvature, we measure the density jump at the point which is tangent to our LOS. The Rankine–Hugoniot shock jump conditions relate the measured density jump to the Mach number of the component of gas velocity traveling perpendicular to the shock surface. Since our LOS is tangent to the point at which we measure this Mach number, the measured velocity component must be in the plane of the sky. We can use this fact, in combination with the LOS velocities of the northern major remnant core and the southern minor remnant core, to constrain the inclination of the merger axis along the LOS and to approximate the shock velocity. To do this, we assume that the southern compact and northern cores have the same LOS velocities as their respective galaxy components, i.e., the southern minor and northern major remnant cores (measured from the three-dimensional KMM partition in Section 4.3). The shock leading the southern compact core is meeting the unshocked gas associated with the northern core with an LOS velocity v_los ≃ 2574 + 1658 = 4232 km s^-1 which, assuming the unshocked gas has kT ∼8.6 keV, gives an LOS Mach number of M_los ≃ 4232/1523 = 2.78. The inclination angle of the velocity vector to our LOS is then arctan(1.41/2.78) ≃ 27° and the total Mach number is M_tot = 3.1 leading to a shock speed of ∼4750 km s^-1, similar to that observed in the Bullet cluster (Markevitch et al. 2002; Markevitch 2006). Using the Mach number derived from the temperature jump gives a similar value for the inclination angle of ∼33°, the total Mach number, M_tot = 3.31 and a shock speed of 5041 km s^-1. We note that, as in the Bullet cluster (Springel & Farrar 2007; Mastropietro & Burkert 2008), the shock velocity is measured relative to the surrounding gas which will have high bulk motions due to the merger. These measurements also rely critically on the assumption that the velocities of the gaseous and galactic components are the same. Future X-ray telescopes, such as the International X-ray Observatory, will have high enough spectral resolution to directly probe the LOS gas kinematics and will enable us to test this assumption.

The measured velocity dispersion of the southern minor remnant core ranges from 581 km s⁻¹, measured from the one-dimensional KMM partition, to 441 km s⁻¹, measured from the three-dimensional KMM partition. This dispersion is likely to be an underestimate of the real dispersion due the KMM partitioning which allocates objects which are probably physically associated with the southern minor remnant core, but in the tails of the velocity distribution, to the main cluster. This effect is seen in the right-most panel of Figure 11, which shows that objects with peculiar velocities overlapping with the main structure have been allocated to the main structure, thereby truncating the southern minor remnant core's velocity distribution. The act of merging will also affect the measured velocity dispersion, since the merger acts to disperse the southern minor remnant core's galaxies (Pinkney et al. 1996). The galaxies most likely to be stripped are those that are least tightly bound to the southern minor remnant core, e.g., those in the outskirts, or those in the tails of the velocity distribution. This stripping, compounded by the lack of certainty in the allocation of galaxies as members of the southern minor remnant core, means it is extremely difficult to gauge the southern minor remnant core's initial velocity dispersion. In any case, the measured velocity dispersion indicates the southern minor remnant core contains significant mass, as does the measured temperature of ∼7 keV for the southern compact core.

Given the above evidence, we interpret the southern compact core as a Bullet-like system (Markevitch et al. 2002) merging along an axis that lies well out of the plane of the sky. To support this interpretation, we present Figure 18, which shows a model for the X-ray appearance of the Bullet cluster (as used in Bradač et al. 2006) viewed from a range of directions. Briefly, the Chandra X-ray image of the Bullet cluster is fitted with a density model which consists of a shuttlecock with concave sides (for the Bullet substructure), power-law profiles across the cold and shock fronts, a flattened pancake (for the main cluster's emission), and a beta model at larger radii. The XIM software (Heinz & Brüggen 2009) is used in combination with this density model to simulate 0.6–5 keV Chandra images with 100 ks exposures and projected so that the merger axis is inclined to our LOS with angles of θ_inc = 90,  15,  35,  48,  and 60. To best match the orientation of the southern compact core in Abell 2744, the images shown in the lower four panels of Figure 18 are again rotated in the plane of the sky.

Zoom In Zoom Out Reset image size

Comparing the simulated images in Figure 18 with the X-ray morphology of the southern compact core, we surmise that an inclination angle in the range θ_inc = 35°–60° provides the best qualitative fit. This is based primarily on the observability of the shock front (Section 2.4.2) and on the blunted cone appearance of the southern compact core in the Chandra images (Figure 1). These inclination angles compare well with that derived above from the combination of the shock and kinematic parameters, particularly when considering uncertainties inherent in both techniques.

In their analysis of the northwestern interloper, KD04 presented tentative evidence for a cold front and a bow shock on the eastern side of the northwestern interloper (their Figure 3). In our deeper observations, we find no convincing evidence for the existence of a shock front at the position and orientation indicated in KD04. However, both of the putative fronts appear to be part of a larger front on the northeastern side of the northwestern interloper, as indicated in Figure 1. The analysis in Section 2.4.3 reveals that this is a cold front. In Figure 2, we noted that the emission surrounding the northwestern interloper extends toward the south. The analyses conducted in Sections 2.3 and 2.4 indicate that this extension is cooler and has lower entropy than the ICM immediately to the east and west. We interpret the low entropy gas as a trail of gas stripped from the northwestern interloper by ram pressure.

While we detect no dynamical system associated with the northwestern interloper, the pair of bright galaxies lying ∼140 kpc and ∼320 kpc southeast of the X-ray peak have peculiar velocities of 561 km s^-1 and 663 km s^-1, respectively. The velocity field (Figure 15) has values of ∼400 km s^-1 in the region of the northwestern interloper, consistent with the velocities of the pair of bright galaxies. This hints at a dynamically distinct system. However, because the velocities are close to those of the main cluster and because of the small number of spectroscopically confirmed cluster members in this region, dynamical substructure is more difficult to detect at a significant level (Pinkney et al. 1996). We note also that if the northwestern interloper has passed pericenter, as we argue in Section 6.2, it is plausible that a large fraction of the galaxies have been tidally stripped. This will further inhibit the detection of dynamical substructure associated with the northwestern interloper.

Using the photometric redshift catalog of Busarello et al. (2002) to define cluster members, probing ∼1.3 mag fainter than our spectroscopy, we do detect a local peak in the galaxy surface density coincident with the northwestern interloper's X-ray peak. We showed that the non-detection of a local peak in the two-dimensional galaxy distribution of the spectroscopically confirmed cluster members is not due to spectroscopic incompleteness, but due to the majority of galaxies associated with the overdensity detected using the photometrically defined members having magnitudes fainter than our limiting magnitude for the spectroscopy. Thus, spectroscopic observations probing to fainter magnitudes are required to confirm the existence of a substructure associated with the northwestern interloper.

The central region of the cluster has extremely complex X-ray morphology, temperature and metallicity structure, as well as multiple dynamically distinct components. These features can be reconciled with an off-center, two-body merger after core passage, with the merger axis nearer to our LOS than the plane of the sky. Below, we provide evidence for these assertions.

During cluster mergers, the strong ram pressure felt by the gas means that it can become decoupled from the collisionless galaxies which will lead the gas (e.g., Barrena et al. 2002; Markevitch et al. 2002; Clowe et al. 2006; Mahdavi et al. 2007; Bradač et al. 2008). From the observed offset between the galaxies and gas, the X-ray morphologies, and also the velocities of the subclusters, we can determine the direction of motion of the two subclusters. The northern major remnant core is blueshifted with respect to the systemic velocity of the cluster complex (v_pec ≃ −1600 km s^-1), the galaxy substructure is offset from the northern core to the northwest, with the BCG associated with the remnant core leading by ∼100 kpc. This evidence indicates that some component of the northern subcluster's motion is directed toward the north–northwest, while a large portion is also directed toward us along the LOS. Further evidence for the north–northwest motion is revealed by the enhanced X-ray emission to the southeast of the northern core (Figures 1 and 17), which we interpret as a trail of stripped gas. The curvature of the gas trail implies significant westward acceleration, so that the northward moving core must have passed to the east of the core of the other major subcluster, which we associate with the southern compact core.

Using a similar argument, we can ascertain the direction of motion of the southern subcluster. The brightest galaxy associated with the southern minor remnant core is offset some ∼140 kpc south of due east of the peak in the X-ray emission associated with the southern compact core and assuming that this bright galaxy was once at the center of the subcluster and coincident with the peak in the X-ray emission, this offset indicates some fraction of the southern subcluster's motion is directed toward the southeast. Further evidence for motion toward the southeast comes from the Chandra image in the form of the edge to the southeast (Figure 1), which is part of the shock front (Section 5.3), and the morphology of the southern compact core, which exhibits a blunted cone structure where the compact core is the head of the blunted cone, and the tail of stripped gas to the northwest forms a wedge-like structure. The mean peculiar velocity of the southern minor remnant core galaxies (∼2300–2500 km s^-1) indicates that a large fraction of the southern substructure's motion is directed away from us along our LOS. However, the edge and blunted cone morphology of the southern compact core show that the direction of motion cannot be entirely directed along our LOS.

These arguments require that the northern and southern subclusters have passed one another and now are moving apart. The gaseous envelopes of the two subclusters have largely been stopped and stripped from their former hosts by shocks, separating them from the remnant cool cores, which can survive the off-center merger due to their higher densities and pressures. In our merger scenario, the central peak of emission is the gas which has been stripped from the outer parts of the two subclusters during the core passage phase of the merger. This interpretation is supported by high temperatures in the central gas, consistent with shock heating and adiabatic compression during the core passage. Patches of low metallicity gas suggest that at least some of this gas originates from outside the more highly enriched cores (Vikhlinin et al. 2005), as described above.

According to the KMM partitioning in Section 4.3, a majority (188 out of 238) of the galaxies are allocated to a partition with μ = −105 km s^-1 and σ = 1438 km s^-1. This is also clearly seen in Figure 10 where within a cluster-centric radius of 500 kpc there is a clear segregation in velocity with essentially no galaxies with v_pec = 0 km s^-1, whereas at R > 500 kpc the velocity distribution appears to be unimodal and centered around v_pec = 0 km s^-1. Presumably, the galaxies at larger radii were once part of the less tightly bound outskirts of the southern and northern substructures. The outskirts of the two merging clusters decoupled from the tightly bound cores during pericenter and are in the process of mixing to form the outskirts of the new cluster.

In our interpretation, the northern subcluster was the more massive of the two merging systems and the southern subcluster is the remnant of the less massive one, consistent with the higher velocity dispersion of the northern major remnant core. However, the southern compact core is brighter now. This suggests that the smaller subcluster hosted a more robust cool core, i.e., a greater mass of low entropy gas, prior to the merger. The higher entropy northern core of the larger subcluster would then be more prone to disruption during the merger. Indeed, the clumpy tail of gas trailing the northern core indicates that a significant fraction of its gas has been removed and it is in the process of being destroyed by the merger.

The deeper Chandra observations presented here fail to confirm the proposal by KD04 that the northwestern interloper is traveling eastward on its first passage through the cluster. KD04 used the positions of two BCGs ∼140 kpc and ∼320 kpc southeast of the northwestern interloper, along with tentative detections of cold and shock fronts on the eastern side of the interloper, to support this scenario. We find little evidence for the shock proposed to exist by KD04, while their proposed cold front is part of a more extended cold front. Furthermore, it is difficult to explain the large separation of gas and galaxies prior to a core passage in a subcluster which appears to have both significant mass (indicated by its temperature kT ∼5 keV) and ICM density (indicated by its X-ray brightness).

Three main features in the deeper X-ray data provide support for an alternative interpretation. They are (1) the cold front to the northeast, (2) the extended region of enhanced X-ray surface brightness to the south, and (3) the swirl of low entropy gas seen in Figure 3, which is associated with the southern extension and is curved in a southeasterly direction toward the cluster center. The cold front to the northeast along with the extension of the emission toward the south indicate that the northwestern interloper is currently traveling toward the north–northeast. The curvature of the swirl of low entropy gas indicates significant angular momentum. Using these features to guide our interpretation, we suggest that the northwestern interloper fell into the main cluster from the south or southeast, initially traveling roughly north or northwest and passing the main cluster core off-center to the southwest. The gravity of the main cluster has subsequently deflected it onto its current path toward the north–northeast (see, e.g., the evolution of the 3:1 and 10:1 mass ratio, offcenter mergers in Poole et al. 2006). The fact these features are observed at all, along with the non-detection of any significant dynamical substructure in this region, indicates the merger of the northwestern interloper is taking place fairly close to the plane of the sky. We note, however, that the non-detection of dynamical activity here may be due to the limited depth and completeness of our sample.

One issue with the above interpretation for the northwestern interloper is the position of the two bright galaxies that are offset to the southeast from the peak in the X-ray emission. Assuming these bright galaxies trace the dominant mass component (i.e., the collisionless dark matter core) of the northwestern subcluster, we would expect to see the classic Bullet-like scenario where the collisionless dark matter and galaxies lead the collisional gas component and thus to see the bright galaxies offset to the northeast of the gas clump. However, we note that during the later stages of a merger, the simulations of Mathis et al. (2005) showed that the cool, dense cores of subclusters can lead the dark matter cores. Indeed, this phenomenon has been observed in the clusters Abell 168 (Hallman & Markevitch 2004) and Abell 754 (Markevitch et al. 2003). The physical mechanism producing this offset is the rapid changes in ram pressure felt by the subcluster gas as it traverses the main cluster; during closest approach, the ram pressure peaks and the gas and dark matter separate, with the dark matter leading. After the core passage phase, the ram pressure drops significantly due to the decrease in the surrounding ICM density and the cool core begins to fall back toward its parent dark matter core, eventually overshooting it in a "ram pressure slingshot" as explained in Markevitch & Vikhlinin (2007). This effect is also simulated in Ascasibar & Markevitch (2006) and their Figure 12 shows qualitative agreement with what is proposed above and with the temperature morphology of the northwestern substructure. High-resolution weak lensing maps are required to determine the location of the local peak in the mass density corresponding to the northwestern interloper and to test the scenario outlined above.

While the northwestern subcluster is due to a more minor merger than the central merger, there should still be observable effects due to the core passage, particularly in the ICM. Although the central major merger complicates this matter and may erase any effects due to the northwestern subcluster, we tentatively point out the bridge of emission in Figure 1 along with the slight asymmetry of the central cluster emission in the direction of the northwestern interloper, as evidence of the previous interaction with the central ICM. We also suggest that the hot regions lying between the main cluster and the northwestern interloper (Figure 3) were produced by shocks and/or compression of the ICM during the subclusters passage.

The interpretations outlined above for the different structures revealed by our new observations give cause to revisit the hypotheses outlined in Boschin et al. (2006) and KD04 for the merging history of Abell 2744. Considering the central regions, KD04 propose a merger scenario where Abell 2744 is seen just prior to a core passage and the merger is occurring along a north–south direction with the majority of the motion directed along our LOS. Boschin et al. (2006), using a slightly larger sample of spectroscopically confirmed cluster members, agreed with KD04 that the central merger is occurring along a north–south direction with the majority of the motion directed along our LOS, but suggested that the merger is in a more advanced post-core-passage phase. Both KD04 and Boschin et al. (2006) suggest that the northwestern subcluster is infalling onto the main cluster from the west and can be treated as separate from the major merger occurring in the central regions.

With regard to the southern subcluster, our interpretation of the new observations is mostly in accordance with the conclusion of Boschin et al. (2006) that it is viewed after a core passage, has a large fraction of its motion directed along the LOS and some component to the south. We slightly modify this scenario based on the detection of the shock front and the blunted cone X-ray morphology, which indicate a more southeasterly direction of motion. The detection of the shock front also indicates that the component of the velocity perpendicular to our LOS must be larger than previously suggested. Indeed, our observations of the shock front, combined with the dynamical analyses, allow us to constrain the direction of motion of the southern subcluster to be inclined at around 30° to the LOS.

Our scenario differs most from that of Boschin et al. (2006) in the importance of the northern subcluster. Boschin et al. (2006) interpret this structure as an additional feature in their scenario due to the lack of a counterpart in the X-ray images. Our deeper Chandra data reveals the northern core as a cool, high metallicity structure which is trailing the dynamically distinct northern major remnant core galaxies. Furthermore, the larger sample of spectroscopically confirmed cluster members provide strong evidence that, when separated by velocity, the most significant projected galaxy overdensity is the northern major remnant core. Taking these new results in concert provides strong evidence in support of our view that the northern subcluster is the surviving core of the main component involved in the central merger. Thus, while we agree with Boschin et al. (2006) that we are observing a post-core-passage merger in the central regions, we suggest a somewhat later phase, given the large (∼800 kpc) projected separation of the main northern subcluster from the high-velocity southern subcluster.

Turning to the interloping northwestern subcluster, the deeper Chandra data have muddied the waters somewhat in terms of a simple interpretation and indicate that the direction of motion of this subcluster is currently to the north/northeast—perpendicular to the direction of motion suggested by both KD04 and Boschin et al. (2006). Both Boschin et al. (2006) and Braglia et al. (2009) detect an overdensity of red galaxies associated with the BCG lying ∼140 kpc to the southeast of the X-ray peak. We propose that this offset has occurred due to a ram pressure slingshot which caused the gas to "overshoot" the galaxies after pericentric passage. This interpretation provides a better fit to the X-ray data and we note that the galaxy overdensities found here and in the Boschin et al. (2006) and Braglia et al. (2009) studies rely on photometric cluster member allocation, rather than the more robust spectroscopic identification of cluster members, which is less prone to projection effects. A more rigorous investigation involving deeper spectroscopy and detailed simulations are required to test the different infall scenarios for this substructure.