Skip to main content

Come analyze HEASARC, IRSA, and MAST data in the cloud! The Fornax Initiative is now welcoming all interested beta users.

Xspec Home Page

next up previous contents
Next: Joint priors: the smoothness Up: Priors Previous: Available prior families   Contents

Practical advice

For most XSPEC users beginning a Bayesian analysis, the following default choices work well:

  • Set jeffreys on every normalization and on any column density whose value crosses multiple decades. Use newpar first to push the parameter's hard lower limit to a small positive value (e.g. $10^{-6}$) if it is currently zero.
  • Set gauss or cauchy on physical parameters for which a published value or theoretical expectation exists, using the published $1\sigma$ uncertainty as the prior width.
  • Leave cons on parameters whose hard limits are already a fair representation of physical possibility (e.g. photon index restricted to $0 < \Gamma < 5$).
  • Switch Bayesian inference on with bayes on once all priors are in place. The samplers will not consult the prior unless this is set.

Hyperparameters — the $h_1, h_2, \dots$ that appear in the prior expressions — should be drawn from prior knowledge or pilot fits, not tuned to the same data the posterior will be conditioned on. Doubly using the data corrupts the credible-interval coverage.