Exact penalty functions in nonlinear programming