Parameter distributions: use appropriate integration rules
Currently, parameter distributions are evaluated at equidistant points. We should get better accuracy by using an established integration rule; maybe we should offer users the choice between different rules.
For distributions with infinite support (Gauss, Lorentz, log-normal), we currently have an extra cutoff parameter (sigma_factor). This we can avoid
- for Gauss distributions, by using Gauss-Hermite integration,
- for Lorentz distributions, by transforming x = tan y,
- for log-normal, Gauss-Hermite with transform x = exp y.
Steps towards resolution:
-
Simplify ScanResolution -
Remove RangedDistribution
Edited by Wuttke, Joachim