Parameter Estimation In X-ray Astronomy Revisited

Abstract

The method of obtaining confidence intervals on a subset of the total number of parameters (p) of a model used for fitting X-ray spectra is to perturb the best-fitting model until, for each parameter, a range is found for which the change in the fit statistic is equal to some critical value. This critical value corresponds to the desired confidence level and is obtained from the chi-square distribution for q degrees of freedom, where q is the number of interesting parameters. With the advent of better energy-resolution detectors, such as those onboard ASCA it has become more common to fit complex models with narrow features, comparable to the instrumental energy resolution. To investigate whether this leads to significant non-Gaussian deviations between data and model, we use simulations based on ASCA data and we show that the method is still valid in such cases. We also investigate the weak-source limit as well as the case of obtaining upper limits on equivalents widths of weak emission lines and find that for all practical purposes the method gives the correct confidence ranges. However, upper limits on emission-line equivalent widths may be over-estimated in the extreme Poisson limit.