Adaptive mesh refinement (AMR) is a design feature of numerical models handling multi-dimensional data such as fields. Examples include qualities spread over three dimensions such as gas density, pressure and flow, temperature, and magnetic fields. Models can make use of a mesh, modeling the values at mesh points, the amount of detail depending upon the mesh scale: how widely spaced are the points recorded and modeled. There may be a resolution (spacing of the mesh) that works for much of the modeling, is not detailed enough for some particular circumstances. AMR is the inclusion of programming logic to move to (and possibly from) the smaller mesh scale in just those regions and times where necessary, based upon detecting those circumstances. The goal is to provide model accuracy and precision while keeping the amount of calculation tractable.
The method, for example, can be used for atmospheric modeling (Earth weather prediction), design of devices involving fluid dynamics (water pipes, rocket engines), and astronomical phenomena including stars, stellar magnetic fields, galaxies, and cosmological simulations of a significant representative portion of the universe.