1Blurb:: 2Use penalty merit function 3Description:: 4Second, the surrogate constraints in the approximate subproblem can be 5selected to be surrogates of the original constraints (\c 6original_constraints) or linearized approximations to the surrogate 7constraints (\c linearized_constraints), or constraints can be omitted 8from the subproblem (\c no_constraints). Following optimization of the 9approximate subproblem, the candidate iterate is evaluated using a 10merit function, which can be selected to be a simple penalty function 11with penalty ramped by SBL iteration number (\c penalty_merit), an 12adaptive penalty function where the penalty ramping may be accelerated 13in order to avoid rejecting good iterates which decrease the 14constraint violation (\c adaptive_penalty_merit), a Lagrangian merit 15function which employs first-order Lagrange multiplier updates (\c 16lagrangian_merit), or an augmented Lagrangian merit function which 17employs both a penalty parameter and zeroth-order Lagrange multiplier 18updates (\c augmented_lagrangian_merit). When an augmented Lagrangian 19is selected for either the subproblem objective or the merit function 20(or both), updating of penalties and multipliers follows the approach 21described in \cite Con00. 22 23Topics:: 24Examples:: 25Theory:: 26Faq:: 27See_Also:: 28