Multiplicity in the Release 2 CSC: Stacks, Ensembles, Bundles and Blocks
To improve the limiting sensitivity of release 2 of the CSC in regions of the sky that have multiple overlapping observations, source detection is performed on co-added observations. This brings some added complexity that is described in detail at various places in the documentation, but we summarize the key concepts here for convenience:
Observations taken with the same instrument (ACIS or HRC-I) and whose pointing direction is co-aligned within 1 arcminute are grouped in stacks. Formally, the observations are matched using a tree clustering algorithm with complete linkage. This means that the pointing direction of each observation in the stack is consigned with the pointing direction of every other observation in the stack within 1 arcminute, and the stacks include the maximum number of observations for which this condition is satisfied.
Source detection is performed on the co-added stacks rather than on the individual observations. These stacks are termed 'detect stacks'. Once a source is detected, per-detection properties are evaluated using the subset of observations for which the source falls on the valid per-observation pixel mask (essentially, the pixels for which the exposure is greater than ~10% of the peak exposure). This subset of observations is termed the 'valid stack' for the detection. See Figure 1.
A complete set of (one or more) stacks that overlap is termed an 'ensemble'. Ensembles are processed together to identify unique ('master') sources and determine source properties.
Groups of detections whose source regions and/or associated background regions overlap are collected into 'bundles'. Detections in a bundle are typically processed together as a set to ensure that a consistent set of properties is determined. For example, aperture photometry is performed on a bundle basis to ensure that photon counts in the overlap regions of multiple detections are accounted for correctly. Bundles are redefined at several stages during the processing.
Aperture photometry measures in release 2 of the CSC are derived using a Bayesian algorithm that computes the flux probability density function for each individual detection (observation) of a source. These measures are then evaluated using a Bayesian Blocks algorithm to group the observations into equivalence classes, called 'blocks', such that for all observations in a single block, the multi-band flux values are consistent with no variation. There are two types of blocks, based on how the Bayesian Blocks algorithm groups the observations. The first are 'flux-ordered blocks', where only the multi-band aperture photometry fluxes are considered. In this way, multiple 'high state' and 'low state' observations of a variable source will be automatically separated from each other, but consistent observations can be combined to improve S/N even if they are separated by observations of the source in another state. The second are 'time-ordered blocks', where in addition only consecutive observations can be grouped together, which is appropriate when computing temporal variability measures. In the catalog tables we report aperture photometry measures (and derived measures such as hardness ratios) for the flux-ordered and time-ordered blocks with the highest exposure time, as well as global average fluxes. Properties for all blocks are included in the source blocks3 FITS data product that can be downloaded via CSCview. See the discussion of the Source Properties pipeline and of relevant table columns and data products for more details.
Catalog Entries: Master source, stack detection and source observation
Because the size of the Chandra PSF varies by roughly two orders of magnitude between the center and edge of the field of view, in the Chandra Source Catalog we differentiate between detections and sources. Detections are the blobs of photon counts that we see on the detector image, whereas sources are our interpretation of the detections as distinct X-ray sources on the sky. There is potentially a many-to-many relationship between detections and sources. A single source may correspond to multiple detections because the source is included in more than one observation that includes the same region of the sky. Similarly, a single detection may correspond to multiple sources; this typically occurs when a single far off-axis detection (large PSF) is resolved into multiple sources by another observation where the same position on the sky is located closer to the telescope optical-axis (small PSF).
Each distinct X-ray source on the sky identified in the Chandra Source Catalog is represented by a single Master Sources Table entry and one or more Stacked Observation Detections Table entries for each observation stack in which the source was located. Each stacked observation detection table entry is accompanied by one or more Per-Observation Detections Table entries with information from each observation making up the stack.
The catalog is modeled as three table views: the Master Sources Table, the Stacked Observation Detections Table and the Per-Observation Detections Table. In the Master Source Table, each row represents a source, and each column contains a quantity or parameter describing that source. In the other two tables, each row represents a detection, and each column contains a quantity or parameter describing that detection. The three tables correspond to properties determined at the ensemble, stack, and individual observation level respectively.
If a source is detected in multiple stacks that overlap the same region of sky, then the master source table entry is produced by combining the corresponding individual stack detection entries. The properties of a master source represent the best estimates of the actual source properties derived from the set of individual source observations and stack detections contributing to the master source observation. The master source properties are recorded in the Master Sources Table.
The Stacked Observation Detections Table contains detection properties based on observational data extracted independently from each stack of Chandra pointings in which the source is detected; because a source may be detected in multiple stacks, this table may contain multiple entries corresponding to a single master source entry in the Master Sources Table.
The Per-Observation Detections Table contains source properties based on observational data extracted independently from each individual observation (Chandra pointing) in which the source is located; it may contain multiple entries corresponding to a single master source entry in the Master Sources Table.
The Column Descriptions pages describe how each master source, stacked observations detection, and per-observation detection property is determined.
Additionally, a number of observation-specific properties are reported in the Per-Observation Detections Table, such as pointing and instrument information, as well as links to file-based full-field data products, such as background maps, bad pixel files, and exposure maps. A number of source-specific data products are produced as well, including source region event and PSF files, among others; see the Data Products section for the full list of CSC file-based data products.
Each individual source detection in the catalog is classified as either uniquely or ambiguously linked to a master source ('match_type' equals 'u' or 'a'). A per-stack source that is uniquely linked to a master source can only match a single source on the sky. The detections properties contribute to the master source properties recorded in the Master Sources Table. A per-stack detection that is ambiguously linked to a master source could match more than one source on the sky. Because the detections photon events cannot be assigned unambiguously to a specific source in the sky, its properties do not contrubute to the properties of any linked master source recorded in the Master Sources Table.
Note that even in the case of uniquely linked per-stack detections, if an ACIS observation is piled-up (estimated pile-up fraction exceeds ~10%), its per-observation properties will not contribute to the corresponding stack or master source properties UNLESS all other ACIS observations of that source are also piled-up.
Figure 2 demonstrates that all "master source/stacked observation detection/per-observation detection associations" in the catalog, whether ambiguous or unique, are transparent to the user. This feature allows a request of any combination of master source properties, stacked detection properties and per-observation detection properties in a single query to the database, so the user is not restricted to searching a single table at a given time, and can understand how all observations of a single source contribute to a master source entry.
New in CSC 2.0 are non-detections, which are linked to a master source with 'match_type' = 'n'. If a master source position falls on a stacked observation for which there is no corresponding detection that can be linked to the master source, then non-detection regions are created at the master source position for each of the observations that comprise that stacked observation. These non-detection regions are sized based on the local PSF (90% ECF), and are used to compute aperture photometry upper limits for the source brightness at the times of each of the observations.
The Catalog Processing section describes in detail how the catalog is created from individual Chandra data sets; see Figure 3 (NOTE: Figure has not yet been updated for Release 2) for a pictorial demonstration of the connections among all components of the catalog.