reachOptions Reference

Complete reference for the reachOptions struct passed to NN.reach(), GNN.reach(), and NNCS reachability methods.

---

All reachability options are passed as fields of a MATLAB struct:

reachOptions = struct;
reachOptions.reachMethod = 'approx-star';
reachOptions.numCores = 1;
output_sets = net.reach(input_set, reachOptions);

Fields

Field	Type	Default	Description
reachMethod	string	'approx-star'	Reachability method: 'exact-star', 'approx-star', 'approx-zono', 'abs-dom', 'direct', 'krylov', 'ode45'
numCores	int	1	Number of parallel workers for computation
relaxFactor	double	0	For approx-star: 0 = full LP (most precise), 1 = no LP (fastest)
lp_solver	string	'linprog'	LP solver backend: 'linprog', 'glpk', 'gurobi'
dis_opt	string	[]	Set to 'display' for verbose debugging output
device	string	'cpu'	Computation device: 'cpu' or 'gpu'

Conformal Prediction Fields

Additional fields for probabilistic verification via verify_robustness_cp:

Field	Type	Default	Description
coverage	double	0.99	Coverage probability (1 - miscoverage level)
confidence	double	0.99	Confidence level for the PAC guarantee
train_mode	string	'Linear'	Surrogate model type: 'Linear' or 'ReLU'
train_device	string	'cpu'	Python training device: 'cpu' or 'gpu'
train_epochs	int	100	Number of surrogate training iterations
train_lr	double	0.001	Surrogate learning rate

NNCS Reach Parameters

For NNCS reach() methods, use a reachPRM struct:

Field	Type	Default	Description
init_set	Star	(required)	Initial state set
ref_input	Star	(required)	Reference input set (e.g., target setpoint)
numSteps	int	(required)	Number of control periods to simulate
reachMethod	string	'approx-star'	Same options as NN reachability
numCores	int	1	Parallel workers