Numerical studies of stochastic approximation procedures for constrained problems

Several algorithms for stochastic approximation under equality and inequality, constraints are discussed and described, together with numerical data and comparisons from numerous simulations. The algorithms work well, and exhibit some rather interesting behavior. Those based on "augmented Lagrangian" techniques are preferable to the "Lagrangian" methods, as in the deterministic case; the former methods seem to be quite robust and reliable. The study is the first (to the authors' knowledge) numerical study of such algorithms.