Struct ProjectedGradient 

pub struct ProjectedGradient;

Gradient projection onto a feasible box.

Computes the projected gradient direction used to detect active constraints and to evaluate optimality of the current iterate.

Implementations

impl ProjectedGradient

pub fn project( x: &[f64], g: &[f64], lower: &[Option<f64>], upper: &[Option<f64>], ) -> Vec<f64>

Compute the projected gradient at x given box bounds.

For each component:

pg[i] = g[i]  if lᵢ < xᵢ < uᵢ  (interior)
pg[i] = min(g[i], 0)  if xᵢ == lᵢ (at lower bound)
pg[i] = max(g[i], 0)  if xᵢ == uᵢ (at upper bound)

pub fn optimality_measure( x: &[f64], g: &[f64], lower: &[Option<f64>], upper: &[Option<f64>], ) -> f64

Compute the optimality measure: ‖x - P(x - g)‖_∞

This is the standard bounded optimality measure used in L-BFGS-B convergence tests.