heur_multistart.h
Go to the documentation of this file.
30 * The heuristic applies multiple NLP local searches to a mixed-integer nonlinear program with, probably nonconvex,
31 * constraints of the form \f$g_j(x) \le 0\f$. The algorithm tries to identify clusters which approximate the boundary
32 * of the feasible set of the continuous relaxation by sampling and improving randomly generated points. For each
33 * cluster we use a local search heuristic to find feasible solutions. The algorithm consists of the following four
38 * Sample random points \f$ x^1, \ldots, x^n \f$ in the box \f$ [\ell,u] \f$. For an unbounded variable \f$ x_i \f$
39 * we consider \f$ [\ell_i,\ell_i + \alpha], [u_i - \alpha,u_i], \f$ or \f$ [-\alpha / 2, \alpha / 2]\f$ for an \f$
44 * For each point \f$ x^i \f$ we use a gradient descent method to reduce the maximum infeasibility. We first compute
56 * where \f$ n_j \f$ is the number of strictly positive \f$ d_j \f$. The algorithm is called Constraint Consensus
62 * We use a greedy algorithm to all of the resulting points of step 3. to find clusters which (hopefully) approximate
67 * Depending on the current setting, we solve a sub-problem for each identified cluster. The default strategy is to
68 * compute a starting point for the sub-NLP heuristic (see @ref heur_subnlp.h) by using a linear combination of the
75 * Since the sub-NLP heuristic requires a starting point which is integer feasible we round each fractional
80 /*---+----1----+----2----+----3----+----4----+----5----+----6----+----7----+----8----+----9----+----0----+----1----+----2*/
Definition: struct_scip.h:68
type definitions for return codes for SCIP methods
type definitions for SCIP's main datastructure
SCIP_RETCODE SCIPincludeHeurMultistart(SCIP *scip)
Definition: heur_multistart.c:1004
common defines and data types used in all packages of SCIP
Definition: objbenders.h:43