Abstract
Many of the existing computations of initial‐ and boundary‐value problems in fluid mechanics suffer from unrealistic treatment of boundary points. Three categories of boundaries are discussed briefly: rigid walls, arbitrary boundaries of a computational region in a subsonic flow, and shock waves. An attempt is made to show in what sense the numerical treatment of such boundaries may be physically wrong and what can be done instead. Examples from the blunt body problem, the transonic flow in a nozzle, the incompressible inviscid flow past a circle, and the quasi‐one‐dimensional flow in a Laval nozzle, are shown.