Distributed source model

A distributed source model replaces the search for a few point dipoles with a dense grid of fixed dipoles — typically thousands of sources tiling the cortical surface, often constrained to be oriented perpendicular to it. The inverse problem then becomes estimating the amplitude of every source simultaneously: a massively underdetermined linear problem that must be regularized by a prior on the source distribution.

Different priors define different methods. An L2 (minimum-energy) prior gives the minimum-norm estimate and its noise-normalized variants; an L1 or sparsity prior gives focal solutions; a spatial-smoothness prior gives LORETA-type solutions. Distributed models avoid pre-specifying the number of sources and yield full cortical maps, but their spatial resolution is limited and biased by the chosen prior, and they typically over-smooth (or, for sparse priors, can produce unstable point solutions).