Presets

Uno implements presets, that is combinations of ingredients and hyperparameters that mimic existing solvers. Uno currently supports:

• filtersqp mimics filterSQP (trust-region feasibility restoration filter SQP method with exact Hessian);
• ipopt mimics IPOPT (line-search feasibility restoration filter barrier method with exact Hessian and primal-dual inertia correction).

The values listed below are taken from uno/options/Presets.cpp.

The default preset auto decides between filtersqp and ipopt, depending on the properties of the problem. The following (simple) oracle is currently used:

• if \(2000 \le n + m\) or \(50000 \le nnz(\nabla c) + nnz(\nabla^2 \mathcal{L})\), pick the ipopt preset
• otherwise, pick the filtersqp preset

filtersqp preset

Option	Type	Value
constraint_relaxation_strategy	string	feasibility_restoration
inequality_handling_method	string	inequality_constrained
hessian_model	string	exact
inertia_correction_strategy	string	none
globalization_mechanism	string	TR
globalization_strategy	string	fletcher_filter_method
filter_type	string	standard
progress_norm	string	L1
residual_norm	string	L2
TR_radius	double	10
l1_constraint_violation_coefficient	double	1
primal_tolerance	double	1e-6
dual_tolerance	double	1e-6
switch_to_optimality_requires_linearized_feasibility	bool	true
protect_actual_reduction_against_roundoff	bool	false

If not provided, the QP solver is chosen automatically from the available QP solvers (if any).

ipopt preset

Option	Type	Value
constraint_relaxation_strategy	string	feasibility_restoration
inequality_handling_method	string	interior_point
barrier_function	string	log
hessian_model	string	exact
inertia_correction_strategy	string	primal_dual
globalization_mechanism	string	LS
globalization_strategy	string	waechter_filter_method
filter_type	string	standard
filter_beta	double	0.99999
filter_gamma	double	1e-8
switching_delta	double	1
filter_ubd	double	1e4
filter_fact	double	1e4
filter_switching_infeasibility_exponent	double	1.1
armijo_decrease_fraction	double	1e-8
LS_backtracking_ratio	double	0.5
LS_min_step_length	double	5e-7
barrier_tau_min	double	0.99
barrier_damping_factor	double	1e-5
l1_constraint_violation_coefficient	double	1000
progress_norm	string	L1
residual_norm	string	INF
primal_tolerance	double	1e-8
dual_tolerance	double	1e-8
loose_primal_tolerance	double	1e-6
loose_dual_tolerance	double	1e-6
loose_tolerance_iteration_threshold	int	15
switch_to_optimality_requires_linearized_feasibility	bool	false
LS_scale_duals_with_step_length	bool	true
protect_actual_reduction_against_roundoff	bool	true

If not provided, the linear solver is chosen automatically from the available linear solvers (if any).