With domination of computation over analysis, issues of convergence and uniqueness have largely given way in the engineering/physics literature to issues of numerical stability. A common cause is sometimes the genesis of a lack of convergence or of uniqueness when seeking an analytical prediction, and of a numerical instability when seeking a numerical prediction. If the underlying cause is the same, the actions prompted are sometimes quite different. The introduction of dissipation is widely accepted as a ‘‘tool’’ for ‘‘regularizing’’ numerical codes, by virtue of incorporating a truer physics into the model. Three experiment scenarios are described for which the introduction of dissipation also serves to ‘‘mask’’ errors in the development behind the code, thereby precluding the prediction of significant physical effects. Two of the scenarios are drawn from the published literature—the errors in modeling arise in including finite frequency effects in the scattering by a rough surface, and in including inclusion/inclusion interaction in estimates of the settling rate for nondilute fluid suspensions. The third scenario has not been discussed in the published literature—the modeling error applies to the generalization of a nonreflecting boundary condition at an interior plane of an unbounded range-independent full space, to now model the reflection that obtains at a plane interface connecting two, dissimilar range-independent half-spaces.
Read full abstract