Parameters

Parameters control the operation of the Cardinal Optimizer. They can be modified before the optimization begins. Each parameter has its own default value and value range. Before starting the solution, user can set the parameters to different values, so as to put forward specific requirements on the solution algorithm and solution process. Of course, the default settings can also be maintained.

According to the task performed by the solver COPT and the optimization problem solved, it can be divided into different types of parameters.

This chapter will introduce the parameters constants provided by COPT and their meanings.

Limits and tolerances

Table 16 Limits and tolerances parameters

Name	Type	Description
TimeLimit	Double	Time limit of the optimization
SolTimeLimit	Double	Time limit if a primal feasible solution has been found
NodeLimit	Integer	Node limit of the optimization
BarIterLimit	Integer	Iteration limit of barrier method
MatrixTol	Double	Input matrix coefficient tolerance
FeasTol	Double	The feasibility tolerance
DualTol	Double	The tolerance for dual solutions and reduced cost
IntTol	Double	The integrality tolerance for variables
RelGap	Double	The relative gap of optimization
AbsGap	Double	The absolute gap of optimization

• TimeLimit

  Double parameter.

  Time limit of the optimization.

  Default: 1e20

  Minimal: 0

  Maximal: 1e20

• SolTimeLimit

  Double parameter.

  Time limit if a primal feasible solution has been found.

  Default: 1e20

  Minimal: 0

  Maximal: 1e20

• NodeLimit

  Integer parameter.

  Node limit of the optimization.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: INT_MAX

• BarIterLimit

  Integer parameter.

  Iteration limit of barrier method.

  Default: 500

  Minimal: 0

  Maximal: INT_MAX

• MatrixTol

  Double parameter.

  Input matrix coefficient tolerance.

  Default: 1e-10

  Minimal: 0

  Maximal: 1e-7

• FeasTol

  Double parameter.

  The feasibility tolerance.

  Default: 1e-6

  Minimal: 1e-9.

  Maximal: 1e-4

• DualTol

  Double parameter.

  The tolerance for dual solutions and reduced cost.

  Default: 1e-6

  Minimal: 1e-9

  Maximal: 1e-4

• IntTol

  Double parameter.

  The integrality tolerance for variables.

  Default: 1e-6

  Minimal: 1e-9

  Maximal: 1e-1

• RelGap

  Double parameter.

  The relative gap of optimization.

  Default: 1e-4

  Minimal: 0

  Maximal: DBL_MAX

• AbsGap

  Double parameter.

  The absolute gap of optimization.

  Default: 1e-6

  Minimal: 0

  Maximal: DBL_MAX

Presolving and scaling

Table 17 Presolving and scaling parameters

Name	Type	Description
Presolve	Integer	Level of presolving before solving a model
Scaling	Integer	Whether to perform scaling before solving a problem
Dualize	Integer	Whether to dualize a problem before solving it

• Presolve

  Integer parameter.

  Level of presolving before solving a model.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off.

  1: Fast.

  2: Normal.

  3: Aggressive.

  4: No Limitations, continues until the model cannot be modified (may be very time-consuming).

• Scaling

  Integer parameter.

  Whether to perform scaling before solving a problem.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No scaling.

  1: Apply scaling.

• Dualize

  Integer parameter.

  Whether to dualize a problem before solving it.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No dualizing.

  1: Dualizing the problem.

Linear programming related

Table 18 Linear programming related parameters

Name	Type	Description
LpMethod	Integer	Method to solve the LP problem
DualPrice	Integer	Specifies the dual simplex pricing algorithm
DualPerturb	Integer	Whether to allow the objective function perturbation when using the dual simplex method
BarHomogeneous	Integer	Whether to use homogeneous self-dual form in barrier
BarOrder	Integer	Ordering algorithm in barrier method
BarStart	Integer	Algorithm for finding initial points in barrier method
Crossover	Integer	Whether to use crossover
ReqFarkasRay	Integer	Advanced topic. Whether to compute the dual Farkas or primal ray when the LP is infeasible or unbounded

• LpMethod

  Integer parameter.

  Method to solve the LP problem.

  Default: -1

  Possible values:

  -1: Choose automatically.

  For Linear Programming, choose dual simplex method;

  For Mixed Integer Linear Programming, choose dual simplex or barrier method.

  1: Dual simplex.

  2: Barrier.

  3: Crossover.

  4: Concurrent (Use simplex and barrier simultaneously).

  5: Choose between simplex and barrier automatically (Based on features such as sparsity and/or coefficients ranges).

  6: First-order method (PDLP).

Note:

Currently, COPT’s GPU mode only supports solving Linear Programming problems using the first-order method (PDLP). To enable it, you need to set LpMethod=6 first.

• DualPrice

  Integer parameter.

  Specifies the dual simplex pricing algorithm.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Using Devex pricing algorithm.

  1: Using dual steepest-edge pricing algorithm.

• DualPerturb

  Integer parameter.

  Whether to allow the objective function perturbation when using the dual simplex method.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No perturbation.

  1: Allow objective function perturbation.

• BarHomogeneous

  Integer parameter.

  Whether to use homogeneous self-dual form in barrier.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No.

  1: Yes.

• BarOrder

  Integer parameter.

  Ordering algorithm in barrier method.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Approximate Minimum Degree (AMD).

  1: Nested Dissection (ND).

• BarStart

  Integer parameter.

  Algorithm for finding initial points in barrier method.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Simple.

  1: Mehrotra.

  2: Modified Mehrotra.

• Crossover

  Integer parameter.

  Whether to use crossover.

  Default: 1

  Possible values:

  -1: Choose automatically.

  Only run crossover when the LP solution is not primal-dual feasible.

  0: No.

  1: Yes.

• ReqFarkasRay

  Integer parameter.

  Advanced topic. Whether to compute the dual Farkas or primal ray when the LP is infeasible or unbounded.

  Default: 0

  Possible values:

  0: No.

  1: Yes.

Semidefinite programming related

• SDPMethod

  Integer parameter.

  Method to solve the SDP problem.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Primal-Dual method.

  1: ADMM method.

  2: Dual method.

Integer programming related

Table 19 Integer programming related parameters

Name	Type	Description
CutLevel	Integer	Level of cutting-planes generation
RootCutLevel	Integer	Level of cutting-planes generation of root node
TreeCutLevel	Integer	Level of cutting-planes generation of search tree
RootCutRounds	Integer	Rounds of cutting-planes generation of root node
NodeCutRounds	Integer	Rounds of cutting-planes generation of search tree node
HeurLevel	Integer	Level of heuristics
RoundingHeurLevel	Integer	Level of rounding heuristics
DivingHeurLevel	Integer	Level of diving heuristics
SubMipHeurLevel	Integer	Level of Sub-MIP heuristics
FAPHeurLevel	Integer	Level of Fix-and-propagate heuristics
StrongBranching	Integer	Level of strong branching
ConflictAnalysis	Integer	Whether to perform conflict analysis
MipStartMode	Integer	Mode of MIP starts
MipStartNodeLimit	Integer	Limit of nodes for MIP start sub-MIPs

• CutLevel

  Integer parameter.

  Level of cutting-planes generation.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• RootCutLevel

  Integer parameter.

  Level of cutting-planes generation of root node.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• TreeCutLevel

  Integer parameter.

  Level of cutting-planes generation of search tree.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• RootCutRounds

  Integer parameter.

  Rounds of cutting-planes generation of root node.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: INT_MAX

• NodeCutRounds

  Integer parameter.

  Rounds of cutting-planes generation of search tree node.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: INT_MAX

• HeurLevel

  Integer parameter.

  Level of heuristics.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• RoundingHeurLevel

  Integer parameter.

  Level of rounding heuristics.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• DivingHeurLevel

  Integer parameter.

  Level of diving heuristics.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• SubMipHeurLevel

  Integer parameter.

  Level of Sub-MIP heuristics.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• FAPHeurLevel

  Integer parameter.

  Level of Fix-and-propagate heuristics.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• StrongBranching

  Integer parameter.

  Level of strong branching.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Off

  1: Fast

  2: Normal

  3: Aggressive

• ConflictAnalysis

  Integer parameter.

  Whether to perform conflict analysis.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No

  1: Yes

• MipStartMode

  Integer parameter.

  Mode of MIP starts.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Do not use any MIP starts.

  1: Only load full and feasible MIP starts.

  2: Only load feasible ones (complete partial solutions by solving subMIPs).

• MipStartNodeLimit

  Integer parameter.

  Limit of nodes for MIP start sub-MIPs.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: INT_MAX

Parallel computing related

Table 20 Parallel computing related parameters

Name	Type	Description
Threads	Integer	Number of threads to use
BarThreads	Integer	Number of threads used by barrier. If value is -1, the thread count is determined by parameter Threads
SimplexThreads	Integer	Number of threads used by dual simplex. If value is -1, the thread count is determined by parameter Threads
CrossoverThreads	Integer	Number of threads used by crossover. If value is -1, the thread count is determined by parameter Threads
MipTasks	Integer	Number of MIP tasks in parallel

• Threads

  Integer parameter.

  Number of threads to use.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: 128

• BarThreads

  Integer parameter.

  Number of threads used by barrier. If value is -1, the thread count is determined by parameter Threads.

  Default: -1

  Minimal: -1

  Maximal: 128

• SimplexThreads

  Integer parameter.

  Number of threads used by dual simplex. If value is -1, the thread count is determined by parameter Threads.

  Default: -1

  Minimal: -1

  Maximal: 128

• CrossoverThreads

  Integer parameter.

  Number of threads used by crossover. If value is -1, the thread count is determined by parameter Threads.

  Default: -1

  Minimal: -1

  Maximal: 128

• MipTasks

  Integer parameter.

  Number of MIP tasks in parallel.

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: 256

GPU computing related

• GPUMode

  Integer parameter.

  Usage mode of the GPU solver.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Force the use of CPU mode.

  1: Utilize NVIDIA GPU.

Notes

1. The current GPU mode of COPT only supports Linear Programming problems, and the First-order method (PDLP) needs to be selected, that is, the GPU mode only works when LpMethod=6 is set.

2. For Windows and Linux-x86 systems, when GPUMode is set to the default value or 1, COPT will attempt to detect whether it can successfully load the CUDA library required for GPU solver and whether there is a GPU supporting device. If both checks succeed, the GPU mode of COPT will be used for computation. If it fails, the CPU mode of COPT will continue to be used. The printed information when COPT is started varies slightly between the two scenarios.

3. For MacOS, Linux-aarch64 and other systems, only the CPU mode of COPT is available. Even if GPUMode is set to 1, COPT will still launch the CPU mode for computation. The printed information during COPT startup differs slightly between the two scenarios.

4. Deploying the CUDA library is not the necessary dependency for running COPT. If LpMethod=6 is not set (First-order method PDLP is not applied), the GPU mode will not be utilized, the installation of the CUDA library does not impact the functionality of COPT.

• GPUDevice

  Integer parameter.

  Utilize the GPU with the specified device ID (in cases where the running machine has multiple GPUs).

  Default: -1 (Choose automatically)

  Minimal: -1 (Choose automatically)

  Maximal: INT_MAX

d .. _PDLPTol:

• PDLPTol

  Double parameter.

  Convergence tolerance for the first-order method (PDLP).

  Default: 1e-6

  Minimal: 1e-12

  Maximal: 1e-4

IIS computation related

• IISMethod

  Integer parameter.

  Method for IIS computation.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Find smaller IIS.

  1: Find IIS quickly.

Feasibility relaxation related

• FeasRelaxMode

  Integer parameter.

  Method for feasibility relaxation.

  Default: 0

  Possible values:

  0: Minimize sum of violations.

  1: Optimize original objective function under minimal sum of violations.

  2: Minimize number of violations.

  3: Optimize original objective function under minimal number of violations.

  4: Minimize sum of squared violations.

  5: Optimize original objective function under minimal sum of squared violations.

Notices:

• Either FeasRelaxMode=4 or FeasRelaxMode=5 does not support MILP but only LP. Other methods support both two types of problems.

Parameter Tuning related

Table 21 Tuner related parameters

Name	Type	Description
TuneTimeLimit	Double	Time limit for parameter tuning
TuneTargetTime	Double	Time target for parameter tuning
TuneTargetRelGap	Double	Optimal relative tolerance target for parameter tuning
TuneMethod	Integer	Method for parameter tuning
TuneMode	Integer	Mode for parameter tuning
TuneMeasure	Integer	Parameter tuning result calculation method
TunePermutes	Integer	Permutations for each trial parameter set
TuneOutputLevel	Integer	Parameter tuning log output intensity

• TuneTimeLimit

  Double parameter.

  Time limit for parameter tuning. If the parameter value is 0, it will automatically set by the solver.

  Default: 0

  Minimal: 0

  Maximal: 1e20

• TuneTargetTime

  Double parameter.

  Time target for parameter tuning.

  Default: 1e-2

  Minimal: 0

  Maximal: DBL_MAX

• TuneTargetRelGap

  Double parameter.

  Optimal relative tolerance target for parameter tuning.

  Default: 1e-4

  Minimal: 0

  Maximal: DBL_MAX

• TuneMethod

  Integer parameter.

  Method for parameter tuning.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Greedy search strategy.

  1: Broader search strategy.

• TuneMode

  Integer parameter.

  Mode for parameter tuning.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Solving time.

  1: Optimal relative tolerance.

  2: Objective function value.

  3: The lower bound of the objective function value.

• TuneMeasure

  Integer parameter.

  Parameter tuning result calculation method.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: Calculate the average value.

  1: Calculate the maximum value.

• TunePermutes

  Integer parameter.

  Permutations for each trial parameter set. If the parameter value is 0, it will automatically set by the solver.

  Default: 0

  Minimal: 0

  Maximal: INT_MAX

• TuneOutputLevel

  Integer parameter.

  Parameter tuning log output level.

  Default: 2

  Possible values:

  0: No output of tuning log.

  1: Output only a summary of the improved parameters.

  2: Output a summary of each tuning attempt.

  3: Output a detailed log of each tuning attempt.

Callback related

• LazyConstraints

  Integer parameter.

  Whether lazy constraints are part of the model.

  Default: -1

  Possible values:

  -1: Choose automatically.

  0: No.

  1: Yes.

Notes

• This parameter only affects MIP.

Other parameters

• Logging

  Integer parameter.

  Whether to print optimization logs.

  Default: 1

  Possible values:

  0: No optimization logs.

  1: Print optimization logs.

• LogToConsole

  Integer parameter.

  Whether to print optimization logs to console.

  Default: 1

  Possible values:

  0: No optimization logs to console.

  1: Print optimization logs to console.

Methods for accessing and setting parameters

In different programming interfaces, the ways to access and set parameters are slightly different. For details, please refer to the corresponding chapters for each programming language API:

• C API: C API Functions: Accessing and setting parameters

• C++ API: C++ API Reference: Parameters

• C# API: C# API Reference: Parameters

• Java API: Java API Reference: Parameters

• Python API: Python API Reference: Parameters