\(\epsilon\)-subgradient projection algorithm (Q1111095)

Using only certain easily computable \(\epsilon\)-subgradients, an implementable convergent algorithm for finding the minimizer of a non- differentiable objective function subject to a finite number of linear constraints in d-dimensional space is given. The particular objective function consists of the pointwise maximum of a finite system of pseudoconvex functions. At each iteration cycle certain projections are computed. The negatives of these directions are feasible directions of strict descent for the objective function. The convergence of the algorithm is proved. The algorithm is also numerically tested.