Noninformative Bayesian priors for large samples based on Shannon information theory

We consider the problem of producing noninformative prior distributions for Bayesian analysis. The definition of "noninformative" adopted here is based on maximizing an intuitively appealing information measure derived from Shannon information theory. Based on large-sample (asymptotic) considerations, we show how the resulting generally intractable optimization problem can be significantly simplified. This differs from the authors' previous work on noninformative priors, which considered finite-samples and showed how a tractable suboptimal solution could be obtained.