Generating Asymptotically Non-Terminating Initial Values for Linear Programs

Abstract: We present the notion of asymptotically non-terminating initial variable values for linear loop programs. Those values are directly associated to initial variable values for which the corresponding program does not terminate. Our theoretical contributions provide us with powerful computational methods for automatically generating sets of asymptotically non-terminating initial variable values. Such sets are represented symbolically and exactly by a semi-linear space, e.g., characterized by conjunctions and disjunctions of linear equalities and inequalities. Moreover, by taking their complements, we obtain a precise under-approximation of the set of inputs for which the program does terminate. We can then reduce the termination problem of linear programs to the emptiness check of a specific set of asymptotically non-terminating initial variable values. Our static input data analysis is not restricted only to programs where the variables are interpreted over the reals. We extend our approach and provide new decidability results for the termination problem of affine integer and rational programs.