Struct SubproblemSolver 

pub struct SubproblemSolver {
    pub max_cg_iter: usize,
    pub cg_tol: f64,
}

Constrained trust-region subproblem solver using projected conjugate gradient.

Solves: min_{p} g^T p + 0.5 p^T B p s.t. ||p|| <= Δ, A p = 0 (projected on constraint manifold)

Uses the Steihaug-Toint projected CG approach.

Fields

max_cg_iter: usize

Maximum CG iterations

cg_tol: f64

CG convergence tolerance

Implementations

impl SubproblemSolver

pub fn new(max_cg_iter: usize, cg_tol: f64) -> Self

Create a new subproblem solver

pub fn solve<BV>( &self, g: &[f64], b_times_v: BV, radius: f64, ) -> (Vec<f64>, bool)

where BV: Fn(&[f64]) -> Vec<f64>,

Solve the trust-region subproblem using Steihaug-Toint PCG.

Returns the step p and whether the trust region boundary was hit.

Arguments

• g - Gradient vector
• b_times_v - Function computing B*v (Hessian-vector product)
• radius - Trust region radius Δ

Trait Implementations

impl Default for SubproblemSolver

fn default() -> Self

Returns the “default value” for a type.