Direct-dual Fejér methods for problems of quadratic programming

The paper deals with the $S$-technology, which reduces convex problems of quadratic programming to the solution of systems of several linear, and one convex, inequalities. A certain variant of the Fejér method is applied to these systems. In particular, the problem of the constructive separability of convex polyhedral sets by a layer of maximal thickness is solved. This algorithm plays an important role in problems of discriminant analysis.