Optimal image reconstruction in radio interferometry

We introduce a method for analyzing radio interferometry data that produces maps that are optimal in the Bayesian sense of maximum posterior probability density, given certain prior assumptions. It is similar to maximum-entropy techniques, but with an exact accounting of the multiplicity instead of the usual approximation involving Stirling's formula. It also incorporates an Occam factor, automatically limiting the effective amount of detail in the map to that justified by the data. We use Gibbs sampling to determine, to any desired degree of accuracy, the multidimensional posterior density distribution. From this we can construct a mean posterior map and other measures of the posterior density, including confidence limits on any well-defined function of the posterior map.