Value used for determining if rows in the LDL’ factorization should be exchanged. A higher value improves stability.
1e-8
progress_tol
Minimum change in objective function to consider it progress
1e-6
cycle_tol
Allowed number of iterations without progress before terminating
10
iter_limit
Maximum number of iterations before terminating
10000
fval_bound
Maximum allowed objective function value. The solver terminates if the dual objective exceeds this value (since it is a lower bound of the optimal value).
1e30
eps_prox
Proximal regularization. A negative value enables automatic regularization for singular problems. A positive forces proximal regularization.
-1e-6
eq_reduction
Equality-reduction policy: -1 disables reduction, 0 selects it automatically, and 1 forces it when eligible. The equality constraints are then eliminated from the problem before it is solved.
0
eta_prox
Tolerance that determines if a fixed point has been reached during proximal-point iterations. A negative value selects an automatic tolerance: 1e-6 for the default dual_tol, otherwise the smaller of 1e-6 and 0.1*dual_tol. A nonnegative value is used directly.
-1
rho_soft
Weight used for soft constraints (higher enables more violations)
1e-6
w_soft
Linear weight used for soft constraints (higher make the resulting penalty function more exact)
0
rel_subopt
Allowed relative suboptimality in branch and bound: nodes that cannot improve on the objective value J of the incumbent by more than abs_subopt + rel_subopt*abs(J) are pruned
0
abs_subopt
Allowed absolute suboptimality in branch and bound (see rel_subopt)
0
sing_tol
Tolerance for checking if the LDL’ factorization is singular
3.7e-11
refactor_tol
Tolerance for refactoring the LDL’ factorization before terminating
1e-9
time_limit
Maximum wall-clock time in seconds before terminating (0 means no limit). If branch and bound reaches it after an integer-feasible solution has been found, the best such solution is returned (exit flag 5).
0
Exit flags
Value
Status
5
Feasible (e.g., BnB terminated early)
4
Optimal (inexact)
2
Soft optimal
1
Optimal
-1
Infeasible
-2
Cycling detected
-3
Unbounded problem
-4
Iteration limit reached
-5
Nonconvex problem
-6
Initial working set overdetermined
-7
Time limit reached
-8
Unsupported problem
Constraint classification
The type of a constraint is classified through an integer value (called sense), where the bits in this integer encode different properties:
Value
Classification
Description
1
Active
The constraint should be imposed as an equality at the start
2
Lower
Determine which of the lower and upper bound is imposed when active
4
Immutable
Disallow the constraint to be activated/deactivated
Either the upper or lower bound should hold with equality
These flags can be combined through addition. For example, if we want to encode equality constraints in our problem we would mark these as active and immutable, that is, with sense 1+4=5.