Last modified: 18 January 2024

URL: https://cxc.cfa.harvard.edu/csc/proc/chs.html

Convex-Hull Pipeline


The convex-hull pipeline performs the equivalent of the compact-source master-match and source-properties pipelines. CSC 2.0 is the first release to include information on "large" X-ray sources—represented by convex hull regions, and with master-source names that end in "X"—and the detection and properties of these sources should be considered an alpha release.

Please note that not all extended X-ray sources are represented by convex hulls. The compact-source pipeline also contains extended sources - that is, those sources whose emission is larger than the Chandra PSF. These "compact" sources are those for which the extended emission is only several times larger than the PSF area, and the convex-hull sources are for larger cases, but there is no distinct cut off between the two cases.

Augmentation of the detection list

The Combine and Detect pipeline provides a list of convex-hull detections per stack, and the Source Validation pipeline has performed manual review of a subset of these detections. The result is a set of convex-hull polygons for each stack (although most stacks do not actually have any such polygons).

The Combine and Detect pipeline is conservative in terms of the contour threshold used to select candidate convex hulls. This implied that some diffuse extended emission was missed by the pipeline. The list of candidate polygons was therefore augmented by a simple algorithm which identifies additional observation-level polygons which enclose "significant" emission.

This algorithm was applied to all 10,382 individual observations included in the catalog. For each observation, the poly3 file constructed by the mkvtbkg algorithm are used to identify polygons which have a contour-level of 5 (this is less than used by the combine-and-detect pipeline), areas larger than 1000 arcsec^2, and enclose more than 500 counts. These numbers were selected to avoid selecting off-axis compact sources while still being sensitive to extended emission. Each additional polygon is converted to a convex-hull representation by combining observation-level hulls into stack-level hulls, and added to the list.


Master Source Candidates

The stack detections are combined together, identifying all those detections from different stacks which overlap, forming potential master sources. Note that the stack-level convex-hull detections are not required to have the same detection band to be combined. The overlapping polygons are combined using a simple weighted-area scheme to identify a proposed outline for the master source. This scheme can occasionally fail, creating multiple polygons or master hulls that overlap, and these cases are flagged for automatic review in the following stage.

A Convex-hull detection from a stack which does not overlap any other stack-level detection is promoted to a potential master-source with no change made to the stack-level polygon.

Proposed master convex-hull source

[Thumbnail image: There are four images of the same X-ray source, in a two by two grid, showing the same features apart from the top-left version, as this observation only covers the top 20 percent of the source. Each image has an orange, slightly translucent, convex polygon, which is the proposed master-hull outline, and was formed by combining the per-stack convex hulls, which are shown as the white dotted lines.]

[Version: full-size]

[Print media version: There are four images of the same X-ray source, in a two by two grid, showing the same features apart from the top-left version, as this observation only covers the top 20 percent of the source. Each image has an orange, slightly translucent, convex polygon, which is the proposed master-hull outline, and was formed by combining the per-stack convex hulls, which are shown as the white dotted lines.]

Proposed master convex-hull source

The four images show the stack event files and stack-level convex-hull sources (white dotted line) that were identified as a potential master convex-hull source. The proposed convex hull for the master source is drawn on all four images as the orange polygon.

The image in the top-left shows a stack which only partially covers the convex-hull source. The color scale used is a linear scale, chosen to highlight the diffuse emission, and is different for each image as the four stacks have significantly-different exposure times. The exposure times are approximately 1, 2, 210, and 130 kilo-seconds respectively, starting at the top left and going left to right, top to bottom. The images use a large bin scale (approximately four arcsecond square pixels) to show the presence of diffuse emission. The "aim point" of the observations in the top two stacks are significantly further away from the extended source than in the bottom two stacks, which—coupled with the significantly shorter observation times—is why the source looks blurrier in these two stacks.

These detections correspond to the CSC 2.0 source 2CXO J181129.7-192541X, which is the supernova remnant SNR G011.2-00.3. See the source using the CSC 2.0 WWT interface.


Quality Assurance

The potential master sources are reviewed manually, looking for two cases:

  • those stack-level polygons which should be deleted,
  • and for those cases where the proposed master-hull shape needs manual adjustment (this includes the small number of cases where the algorithm failed to generate a valid hull).

The result is a list of master-source convex-hull polygons, along with a list of stacks in which each master-source was detected.


Source properties

As with the compact sources, a number of properties of the convex-hull sources are calculates. The number is significantly less than the compact case, however, as there is no bayesian-block analysis, spectral or timing analysis, and photometry is limited to the aperture-corrected net flux columns. The Convex-Hull properties page contains a list of the master-source properties (i.e. columns) provided for the convex-hull sources.

The source position is calculated using a weighted centroid of the flux distribution from each stack in which the source was detected, and errors are calculated from the maximum of the second moments of the flux distribution along the X and Y directions, and then a correction factor of 2.447747 to convert to an approximate 95% error. An additional factor of 0.70 arcseconds is added in quadrature to account for astrometric uncertainty. As this is considered a circular error, the major and minor axis values are set to the same and the position angle set to 0. The source position is calculated from the broad-band data for ACIS stacks and wide-band data from HRC stacks.

The source name is generated from the position, and matches the same scheme as used for compact sources except that a 'X' character is added to the end of the name. This means that the following "Search Criteria" can be used in CSCview to identify convex-hull sources (where name refers to the master-source name):

name LIKE %X

Alternatively, the search can use the stack extent-code field, which is set to 256 for convex-hull sources (this is under the "Stacked Observation Detections" section in CSCview, not the "Per-Observation Detections" version):

s.extent_code = 256

Fluxes are calculated in a similiar manner to the compact case, in that a background region just larger than the source is created, and regions which contain compact sources are removed, but this time the source region is the convex-hull polygon rather than an ellipse. Fluxes are calculated for each band appropriate for the observing instrument (wide for HRC and soft, medium, hard, and broad for ACIS), and follows the approach taken in the compact-source pipeline. For each observation that contributes to a convex-hull source, the probability density function is calculated, and these are used to create the per-observation value. Simultaneous fits of this data is used to calculate the per-stack and per-master fluxes. Unlike the compact source case, the only fluxes that are calculated are the flux_aper_avg and photflux_aper_avg versions.