user_prior_sigmas

ell1fit.priors.user_prior_sigmas(fit_parameter_names, parameters_with_unc, user_priors)

Report the standard deviation of each overridden prior, for parameter scaling.

The local coordinates the sampler works in – and hence the width of the ball of starting walkers – come from a per-parameter uncertainty estimate (ell1fit.scaling.get_factors()) that knows nothing about the priors. An override narrower than that estimate would start most walkers outside its own support, where the log-posterior is -inf and nothing can move. Reporting the prior’s own width closes that gap.

A uniform’s standard deviation is its width over sqrt(12), not its half-width: a factor 1.7, which mostly disappears into the order-of-magnitude rounding downstream, but only one of the two is a sigma.

Returns:

dict

{name: sigma} for the fitted parameters that carry an override. Parameters without one are absent, not NaN, so the caller can tell “no opinion” from “an opinion worth nothing”.