MOPS Parameter Options

List of Options

Description of Options

Bound Treatment

Restriction of bounds. Dual simplex needs restricted bounds for best performance. The default is set to 0 if IPM or primal simplex is used and to 1 if dual simplex is used.

Pricing

Sets scaling of Reduced Costs.

LU Factorization Tolerance

An element in a row or column will be set to 0 during the LU Factorization, if this value multiplied by a value of an element with the largest magnitude of a row or column.

Zero Tolerance

A matrix coefficient with a smaller magnitude than xdropm will be set to 0.0.

LP Preprocessing Fill

Determines how much fill is tolerated when using the aggregator during LP-preprocessing. If LP Preprocessing Fill is set to a higher value more fill is allowed. As a consequence the preprocessed matrix and its factorizations may have more nonzeros but fewer constraints.

Refactorization Frequency

A new LU-factorization in one of the simplex engines is computed after the value of Refactorization Frequency unless there are numerical problems or a significant amount of fill which makes it more efficient to recompute the LU factors.

Cholesky Memory

Specifies how the memory for the Cholesky factorization of the IPM-engine is allocated. 0 means it is allocated form the MOPS memory block. 1 means the memory is allocated dynamically outside of the MOPS memory block. Note, that the value of Cholesky Memory is only relevant if the IPM engine is used to solve the initial LP. Furthermore if xipmen is set to 1 the size of the MOPS memory block specified by xmreal will be reduced if it is set too large.

Crossover Level

Sets the crossover level when solving LPs with IPM.

LP Method PercentValue

specifies a percent value. If the ratio of the number of active rows and columns after LP-preprocessing is larger than LP Method PercentValue then the IPM-engine solves the dual LP.

Algorithm

Sets the algorithm used to solve the initial LP.

Refactorization Level

A new LU Factorization will be calculated if the number of elements of the current LU Factorization divided by the number of elements of the last Factorization is greater than Refactorization Level.

Threads

Specifies how many processors should be used for the execution of the IPM-engine on a multiprocessor system. For example on an Intel Core Duo architecture Threads should be set to 2, resulting in a 40% reduction of the CPU-time. Note, that the value of Threads is only relevant if the IPM engine is used to solve the initial LP. Note, CP-time will increase if you specify Threads > 1 and the machine has less than xnproc processors.

Cholesky Ordering

Ordering strategy for the Cholesky factorization using the interior point method

Presolve Methods

Presolve action control for LP Preprocessing. The bits of this value represent turning on or off individual presolve techniques. To get a particular form of the presolve, set xpreme to the sum of the values corresponding to the following presolve methods:

Use Preprocessing

Determines if preprocessing will be run.

MIP Absolute Gap

Absolute gap. The Branch and Bound search will be stopped if the best difference between the best MIP solution found so far and the best MIP bound is less than Absolute Gap.

Maximum Extra Columns

Maximum number of additional columns for extending the LP/IP model.

Maximum Extra NZs

Maximum number of additional nonzeros for LP and IP preprocessing.

Maximum Extra Rows

Maximum number of additional constraints for LP and IP preprocessing.

Dual Simplex Bounding

The dual simplex engine benefits from tight bounds derived during the LP-preprocessing. Dual Simplex Bounding determines the level of bound reduction and bound retaining after LP-preprocessing.

Bound Reduction Level

Level of bound reduction during supernode processing.

Branching Strategy

Branching heuristic. Determines how a branching variable will be chosen.

Use Clique Cuts

Determines if Cliques Cuts will be produced during Supernode Processing.

Coefficient Reduction

Sets coefficient reduction level during supernode processing.

Use CoverCuts

Determines if and how Cover Cuts are derived at Supernodes and during branch-and-cut.

Use Gomory Cuts

Determines whether Gomory Mixed-Integer Cuts will be derived during Supernode Processing

Use Euclidean Reduction

Euclidean reduction of constraints during supernode processing.. The default is 1 meaning Euclidean reduction is used.

Flow Cuts

This parameter determines the level of flow cuts that should be derived.

Halt After X nodes

Determines if the Branch-and-Bound cut process is interrupted after reaching this node count. Default is -1 means no interruption in the Branch and Bound process.

MIP Relative Gap

Relative global gap. The Branch-and-Cut search will be stopped if the relative difference between the objective function value of the best integer solution and the objective function value of the best waiting node is less than MIP Relative Gap.

Gomory Density Level

Sets the density level for taking Gomory cuts into the matrix. A violated Gomory cut will only be added to the matrix if the sum of all nonzeros in all columns of the cut is smaller than Gomory Density Level * xn. The smaller value, the more restrictive is the use of cuts.

MIP Heuristic

Sets the IP heuristic to be used for the initial solution.

Use Basic Fixing

Determine whether a part of the none basics are fixed before the heuristic will start. The default is on.

Heuristic Termination

If the global gap in the heuristic is smaller than Heuristic Termination, the heuristic will be terminated. This parameter is only used if a heuristic is used.

Down Integer FeasTol

A fractional integer variable will be rounded down, if its fractional part is smaller than this tolerance parameter. This parameter is valid for the RINS- and RSS-heuristic.

Up Integer FeasTol

A fractional integer variable will be rounded up, if its fractional part is greater than this tolerance parameter. This parameter is valid for the RINS- and RSS-heuristic.

Heuristic Time Limit

Time limit in heuristic and is only relevant if a heuristic is used.

Bound Implication Usage

Determines the use of conditional bound Implications (only mixed integer models).

Deriving Implications

Determines how implications, that have been derived by probing, will be used during Supernode Processing.

LP Method At Nodes

Sets the algorithm to be used to solve the LP at each node during the Branch-and-Bound process. Default is 1 means Dual Simplex will be used, 0 states Primal Simplex will be used.

LIFO Termination

Set type of mixed LIFO. Mixed LIFO is activated if Node Selection < 0.

Local Search Type

Set type of mixed LIFO. Mixed LIFO is activated if Node Selection < 0.

Use MIR Cuts

This parameter determines if mixed integer rounding cuts should be derived during the super node processing.

Use MIR Cuts

This parameter determines if mixed integer rounding cuts should be derived during the super node processing.

Node Heuristic Limit

Sets number of nodes to be processed by the heuristics before the Branch-and-Bound process. This parameter is only relevant if heuristics are used.

Max Disk TreeSize

Maximum disk space in MB used for the node file during the Branch-and-Bound / Cut process.

Branch N Cut Limit

The branch-and-cut search is terminated when the value of Branch N Cut Limit is reached independent of any other search limitation such as CP-time (Branch N Cut Time Limit), node limit (Branch N Cut Node Limit), disk space or global gap(MIP Relative Gap).

Branch N Cut Time Limit

Maximum CPU time in minutes for the initial LP and the Branch-and-Bound process. The IP optimization will be stopped if xmxmin has been reached.

Branch N Cut Node Limit

Maximum number of nodes in Branch-and-Bound process. The IP optimization will be stopped after Branch N Cut Node Limit nodes.

Max Processing Passes

Max number of passes in Supernode Processing for improving the strength of the LP Relaxation. The Supernode Processing will be terminated after Max Processing Passes passes or if the objective function value does not change anymore.

Select Brand N Bound Alg

Indicate whether new branch and bound algorithm is executed. Default is 0, thus old Branch N bound used.

Local Search Level

Node limit in branch and bound before the next local search or mixed LIFO iteration starts. This parameter is only used in the new branch and bound and is only relevant if Node Selection < 0 or Local Search Type > 0.

Node Limit Local Search

Node limit during local search heuristic or mixed lifo strategy. This parameter is only used in the new branch and bound and is only relevant if Node Selection < 0 or Local Search Type > 0.

Node Selection

Sets the node selection strategy in the Branch-and-Bound process.

MIP Recomputation

Min MIP Solution Improve and MIP Relative Gap are set to default values of 1.d-4. If the IP-model has very large objective function values then it is possible that an optimal IP-solution cannot be found because Min MIP Solution Improve and MIP Relative Gap are relative values. These values are therefore recomputed after the initial supernode processing resulting possibly in smaller values. In such a case they replace the values of Min MIP Solution Improve and MIP Relative Gap. To prevent this recomputation and possible replacement xnogap can be set to 1.

Build Integer Table

Indicate whether integer table is build. The default is off.

Heuristic/LIFO Node Limit

Node limit during local search heuristic or mixed lifo strategy and is only relevant if Node Selection < 0 or Local Search Type > 0.

Node Table Partitioning

Set the partitioning type for the node table.

Partitioning Node Limit

Node limit for a partitioning update, if Node Table Partitioning <> 0.

Initialize PseudoCosts

Set the initialization type for pseudo costs.

Probing

Probing level. Probing tries to set binary variables, that are not fixed, to 0 or 1 and analyzes the resulting implications. In pure 0/1 models this technique can be very effective. In mixed integer models, probing is usually less efficient.

Range Con Conversion

Sets conversion of Range constraints that contain integer variables. The default is off.

Min MIP Solution Improve

After each IP solution found, The current upper bound will be set to best IP solution found - Min MIP Solution Improve * |best IP solution found| * xsscal, so the next integer solution must be Min MIP Solution Improve * 100% better than the previously found solution with objective function value best IP solution found. After the Branch-and-Bound process it will be proved, that the best solution found is maximal Min MIP Solution Improve * 100% away from the global optimum.

Output Level

Output level for logging. Controls creation of *.log and *.sta files.

LP Time Limit

Maximum CPU time in minutes for solving the initial LP. The optimization will be terminated after LP Time Limit minutes if no solution has been found until then.