NMath Reference Guide

## Interior |

The InteriorPointQPSolverParams type exposes the following members.

Properties

Name | Description | |
---|---|---|

KktForm | Controls how the interior point solver computes search directions. Interior point solvers iteratively solve an optimization problem by computing search directions that roughly follow a central path to the solution. Search directions are typically formed by solving a large, sparse linear system. In practice there are different linear systems that result in essentially the same search directions. The KktForm property specifies which variety applies. Sometimes the solver obtains better performance on large problems with the Augmented setting, depending on problem structure. The default is Blended. | |

MaxDenseColumnRatio | The MaxDenseColumnRatio property is the threshold for whether a column is regarded as dense. Dense columns are handled specially by the solver. Default is 0.8. | |

MaxIterations | controls the maximum number of iterations performed by the solver. If the number of iterations is exceeded the solver status will be Interrupted. MaxIterations is less than zero implies no limit on the number of iterations. This is the default. | |

PresolveLevel | Presolve rules eliminate rows or variables from a problem based on simple logical rules. For many problems this can reduce the size and complexity of a problem at minimal cost. For most problems it is best to leave at the default value of Automatic. | |

SymbolicOrdering | Linear systems can be solved more efficiently when the rows and columns are permuted according to a symbolic ordering. The interior point solver currently supports two symbolic ordering techniques: approximate minimum fill and approximate minimum degree. The former generally obtains better orderings but takes longer to run. The default is to let the solver chose the orderng automatically. | |

Tolerance | The interior point solver iteratively improves the infeasibility of the original ("primal") problem as well as the infeasibility of the related "dual" problem. It also monitors the gap between primal and dual iterates. The SolveTolerance property controls the accuracy of each of these values. Consider changing solver tolerance to a larger value if you do not require a high accuracy solution (the default is 1e-8). |

See Also