Struct FiniteDiffHessian 

pub struct FiniteDiffHessian<F>
where
    F: Fn(&ArrayView1<'_, f64>) -> f64 + Send + Sync,
{
    pub fun: F,
    pub x: Vec<f64>,
    pub step: f64,
    /* private fields */
}

Finite-difference full Hessian approximation.

Recomputes the Hessian from scratch at each update call using central differences. Suitable only for small problems (n ≤ a few hundred) because cost is O(n²) function evaluations.

Fields

fun: F

Objective function

x: Vec<f64>

Current iterate x

step: f64

Finite-difference step size

Implementations

impl<F> FiniteDiffHessian<F>

where F: Fn(&ArrayView1<'_, f64>) -> f64 + Send + Sync,

pub fn new(fun: F, x0: &[f64], step: f64) -> Result<Self, OptimizeError>

Create and compute the initial Hessian at x0.

pub fn with_default_step(fun: F, x0: &[f64]) -> Result<Self, OptimizeError>

Create with the default step size.

pub fn recompute_at(&mut self, x: &[f64]) -> Result<(), OptimizeError>

Recompute the Hessian at a new point.

Trait Implementations

impl<F> HessianApproximation for FiniteDiffHessian<F>

where F: Fn(&ArrayView1<'_, f64>) -> f64 + Send + Sync,

fn update(&mut self, _s: &[f64], _y: &[f64]) -> Result<(), OptimizeError>

Update the approximation given a new step-curvature pair.

fn multiply(&self, v: &[f64]) -> Result<Vec<f64>, OptimizeError>

Compute H v (Hessian times vector).

fn inverse_multiply(&self, v: &[f64]) -> Result<Vec<f64>, OptimizeError>

Compute H⁻¹ v (inverse Hessian times vector).

fn to_dense(&self) -> Array2<f64>

Return the current approximation as a dense matrix (Hessian or inverse, implementation-defined).

fn reset(&mut self)

Reset to the initial (identity) approximation.

fn dim(&self) -> usize

Problem dimension n.