Solvers

Recommended Configurations

The following configurations can be seen as helpful starting point on how to configure different solvers for large-scale models. They are largely based on other model’s defaults (see e.g. PyPSA).

More information can be found at:

• HiGHS:

  • HiGHS Documentation - List of Options

• Gurobi:

  • Gurobi Days Paris Advanced Gurobi Algorithms

  • Gurobi Documentation - Parameter Reference

• CPLEX:

  • IBM ILOG CPLEX Optimization Studio - List of CPLEX parameters

  • IBM ILOG CPLEX Optimization Studio CPLEX Parameters Reference (outdated)

HiGHS

solver:
  name: highs
  attributes:
    threads: 4
    solver: "ipm"
    run_crossover: "off"
    small_matrix_value: 1e-6
    large_matrix_value: 1e9
    primal_feasibility_tolerance: 1e-5
    dual_feasibility_tolerance: 1e-5
    ipm_optimality_tolerance: 1e-4
    parallel: "on"
    random_seed: 1234

Gurobi

solver:
  name: gurobi
  attributes:
    Method: 2
    Crossover: 0
    BarConvTol: 1.e-6
    Seed: 123
    AggFill: 0
    PreDual: 0
    GURO_PAR_BARDENSETHRESH: 200
    Threads: 8
    Seed: 1234

Gurobi (NumFocus)

For models with “challenging” numerical properties, the following can be useful:

solver:
  name: gurobi
  attributes:
    NumericFocus: 3
    Method: 2
    Crossover: 0
    BarHomogeneous: 1
    BarConvTol: 1.e-5
    FeasibilityTol: 1.e-4
    OptimalityTol: 1.e-4
    ObjScale: -0.5
    Threads: 8
    Seed: 1234

Gurobi (fallback)

solver:
  name: gurobi
  attributes:
    Crossover: 0
    Method: 2
    BarHomogeneous: 1
    BarConvTol: 1.e-5
    FeasibilityTol: 1.e-5
    OptimalityTol: 1.e-5
    Threads: 8
    Seed: 1234

CPLEX

solver:
  name: cplex
  attributes:
    CPXPARAM_Threads: 8                 # Number of Threads
    CPXPARAM_LPMethod: 4                # Barrier
    CPXPARAM_SolutionType: 2            # No Crossover
    CPXPARAM_Barrier_ConvergeTol: 1e-5
    CPXPARAM_Feasopt_Tolerance: 1.e-5