Solver Options

All options are set via SolverOptions. Defaults match Ipopt where applicable.

#![allow(unused)]
fn main() {
let opts = SolverOptions {
    tol: 1e-8,
    max_iter: 500,
    ..SolverOptions::default()
};
}

Convergence

Option	Default	Description
`tol`	`1e-8`	KKT optimality tolerance (dual infeasibility + complementarity)
`constr_viol_tol`	`1e-4`	Constraint violation tolerance
`dual_inf_tol`	`1.0`	Dual infeasibility tolerance (scaled)
`compl_inf_tol`	`1e-4`	Complementarity tolerance
`max_iter`	`3000`	Maximum iterations
`max_wall_time`	`0.0`	Wall-clock limit in seconds (0 = unlimited)
`stall_iter_limit`	`30`	Max iters without 1% improvement before stall detection (0 = off)
`early_stall_timeout`	`120.0`	Max seconds for first 3 iterations (0 = off)

Barrier parameter

Option	Default	Description
`mu_init`	`0.1`	Initial barrier parameter
`mu_min`	`1e-11`	Minimum barrier parameter floor
`mu_strategy_adaptive`	`true`	`true` = oracle-based (Free mode); `false` = monotone
`kappa`	`10.0`	Barrier decrease divisor: `μ_new = avg_compl / kappa`
`mu_linear_decrease_factor`	`0.2`	Monotone mode: `μ_new = factor * μ`
`mu_superlinear_decrease_power`	`1.5`	Monotone mode superlinear exponent
`barrier_tol_factor`	`10.0`	Subproblem tol = `barrier_tol_factor * μ`
`mu_allow_increase`	`true`	Allow μ to increase after restoration/stall recovery
`adaptive_mu_monotone_init_factor`	`0.8`	Initial μ factor when entering Fixed (monotone) mode
`mu_oracle_quality_function`	`false`	Use quality function for mu selection

Linear solver

Option	Default	Description
`sparse_threshold`	`110`	Switch to sparse multifrontal solver when `n+m ≥ threshold`
`linear_solver`	`Direct`	`Direct` (MUMPS/BK), `Iterative` (MINRES), or `Hybrid`

Inertia-free curvature test (IFRd)

Opt-in alternative to inertia-based regularization (IBR). When the linear solver reports wrong inertia, run the curvature test of Chiang & Zavala (2016, COAP 64:327-354, eq. 28) on the computed direction; accept if the curvature condition holds, otherwise fall back to the standard δ-escalation ladder. Mirrors Ipopt 3.14's IpPDFullSpaceSolver dispatch.

Option	Default	Description
`neg_curv_test_tol`	`0.0`	Curvature acceptance tolerance α_d. `0.0` disables IFRd (pure IBR). Set to `1e-12` to enable.
`neg_curv_test_reg`	`true`	Include δ_w·‖(dx,ds)‖² regularization in the curvature sum (matches Ipopt)

Empirically (CUTEst sweep, 727 problems): default vs. tol=1e-12 both solve 541 to Optimal, but the mix differs — 19 problems are rescued by IFRd and 19 different problems regress. Default off; enable on a per-problem basis when a problem doesn't solve with IBR.

Mehrotra predictor-corrector

Option	Default	Description
`mehrotra_pc`	`true`	Enable Mehrotra predictor-corrector (20–40% fewer iterations)
`gondzio_mcc_max`	`3`	Max Gondzio centrality corrections per iteration (0 = off)

Line search and corrections

Option	Default	Description
`max_soc`	`4`	Max second-order corrections per step
`tau_min`	`0.99`	Fraction-to-boundary parameter (τ in step size rule)
`constraint_slack_barrier`	`true`	Include constraint slack log-barriers in filter merit function
`watchdog_shortened_iter_trigger`	`10`	Consecutive short steps before watchdog
`watchdog_trial_iter_max`	`3`	Watchdog trial iterations

Hessian strategy

Option	Default	Description
`hessian_approximation_lbfgs`	`false`	Use L-BFGS instead of exact Hessian (no `hessian_values` needed)
`enable_lbfgs_hessian_fallback`	`true`	Auto-retry with L-BFGS if exact Hessian IPM fails

Fallback cascade

Option	Default	Description
`enable_slack_fallback`	`true`	Reformulate inequalities with explicit slacks on failure
`enable_al_fallback`	`true`	Try Augmented Lagrangian if IPM fails (equality problems)
`enable_sqp_fallback`	`true`	Try SQP if AL/slack also fail
`enable_lbfgs_fallback`	`true`	Try L-BFGS for unconstrained problems on IPM failure

Restoration

Option	Default	Description
`restoration_max_iter`	`200`	Max iterations for NLP restoration subproblem
`disable_nlp_restoration`	`false`	Disable NLP restoration (prevents recursion in inner solves)

Warm start

Option	Default	Description
`warm_start`	`false`	Initialize from a previous solution
`warm_start_bound_push`	`1e-3`	Bound push for warm-started variables
`warm_start_bound_frac`	`1e-3`	Bound fraction for warm-started variables
`warm_start_mult_bound_push`	`1e-3`	Multiplier floor for warm start

Initial point

Option	Default	Description
`bound_push`	`1e-2`	κ₁: push initial x away from bounds by max(κ₁, κ₂·(u−l))
`bound_frac`	`1e-2`	κ₂: fraction of bound gap used for push
`slack_bound_push`	`1e-2`	Slack variable bound push
`slack_bound_frac`	`1e-2`	Slack variable bound fraction
`least_squares_mult_init`	`true`	Initialize y by least-squares on stationarity
`constr_mult_init_max`	`1000.0`	Cap on LS multiplier init magnitude

Bound thresholds

Option	Default	Description
`nlp_lower_bound_inf`	`-1e20`	Treat variable/constraint bounds below this as -∞
`nlp_upper_bound_inf`	`1e20`	Treat variable/constraint bounds above this as +∞

Preprocessing and detection

Option	Default	Description
`enable_preprocessing`	`true`	Auxiliary equality-block preprocessing/recovery, fixed-variable elimination, and redundant-constraint removal
`auxiliary_tol`	`1e-8`	Accepted residual for auxiliary preprocessing/recovery block solves
`detect_linear_constraints`	`true`	Skip Hessian for constraints with zero second derivatives
`proactive_infeasibility_detection`	`false`	Early infeasibility detection during iterations

Diagnostics

Option	Default	Description
`print_level`	`5`	Verbosity: 0 = silent, 5 = full iteration table + diagnostics

KKT matrix dump (instrumentation)

Option	Default	Description
`kkt_dump_dir`	`None`	If set, write each KKT matrix to this directory after factorization
`kkt_dump_name`	`"problem"`	Problem name prefix for dump filenames

When kkt_dump_dir is set, ripopt writes two files per iteration:

<name>_<iter:04>.mtx — Matrix Market symmetric format, lower triangle
<name>_<iter:04>.json — Metadata: n, m, rhs, inertia, status

This is useful for collecting benchmark matrices for external sparse solvers.