Last modified: 18 January 2024

URL: https://cxc.cfa.harvard.edu/csc/why/err_ellipse_cio.html

Position Errors in the Stacked-Observation Detections Table


Detection Position Error Estimates

In the Stacked-Observation Detections Table, compact detection positions (ra, dec) are determined from the best fitting Maximum Likelihood Estimator (MLE) model. MLE uses Sherpa to perform a simultaneous fit to the set of individual observations that comprise the stack, using the local PSF model for each observation convolved with a rotated elliptical Gaussian. For the point source model, the Gaussian sigmas are set to approximate a single-pixel delta function, whereas for the extended compact source model the sigmas are allowed to vary.

After the Sherpa model fit is performed, the confidence intervals for the position parameters are calculated. The Sherpa implementation of the pyBLoCXS algorithm (Van Dyk et al. 2001, Siemiginowska et al. 2011) is used to obtain a MCMC sample from the posterior probability density for all the fitted parameters. The starting point for the MCMC run is the best fit (maximum likelihood) model parameter values, and the assumed priors for the position parameters are flat within the boundary of the fitted regions.

A sample of 5,000 draws from the posterior probability distribution for a point source model, and 15,000 draws for an extended compact source model. The convergence and correlations within the draws for each parameter are calculated using the \(\hat{R}\) diagnostic parameter (Gelman et al. 2013). In calculations of \(\hat{R}\) the full MCMC chain is split into 10 sub-sections and the variance within each section and between the sections is used as input to \(\hat{R}\) calculation. We verify that the \(\hat{R}\) value for each parameter is close to 1.0 for detections classified as TRUE based on their likelihood. In marginal detections this value can be larger, and this can be due to an incorrect model being used, potential multi-source regions fit with single source, additional emission components not taken into account in the model etc.

Provided that \(\hat{R} < 1.2\) and the acceptance fraction \(> 0.2\), the 90% confidence best-fitting error ellipse is calculated using the MCMC posterior probability draws. If these criteria are not met, then an estimate of the 90% confidence error circle is estimates from a set of position uncertainty equations calibrated from the equivalent circular radii of a large sample of good-quality best-fitting error ellipses. The circular error position uncertainty equation takes the form:

\[ \log_{10}\left(error\ radius \right) = C_{0} + C_{1} \theta + C_{2} \log_{10}\left(net\ counts \right) + C_{3} \theta \log_{10}\left(net\ counts \right) \]

where the constants \(C_{n}\) are listed in the table below.

Energy Band \(C_{0}\) \(C_{1}\) \(C_{2}\) \(C_{3}\)
b -0.031 0.173 -0.526 -0.023
h 0.054 0.176 -0.607 -0.021
m 0.056 0.154 -0.578 -0.018
s 0.088 0.153 -0.582 -0.018
u 0.242 0.134 -0.714 0.001
w 0.075 0.206 -0.544 -0.026
[NOTE]
Note

Position errors in the Stacked-Observation Detections Table are internal errors (i.e., calculated from the model fit to the data) only. They do not include uncertainties in the absolute position of the Chandra data relative to the ICRS astrometric reference frame. Such uncertainties are included in the source position error estimates included in the Master Sources Table.