Struct RiemannianLBFGS 

pub struct RiemannianLBFGS {
    pub history: usize,
    pub step_size: f64,
    pub max_iter: usize,
    pub grad_tol: f64,
}

Fields

history: usize

step_size: f64

max_iter: usize

grad_tol: f64

Implementations

impl RiemannianLBFGS

pub fn minimize( &self, manifold: &dyn RiemannianManifold, objective: &mut dyn RiemannianObjective, initial: ArrayView1<'_, f64>, ) -> GeometryResult<Array1<f64>>

Riemannian L-BFGS with a backtracking-and-expansion Armijo line search.

The search starts at the user-supplied step_size (a hint, not a hard cap) and first expands by doubling while the Armijo sufficient- decrease condition continues to hold and the objective is still strictly improving. Once expansion stalls, it accepts the best step it has seen so far; if even the initial trial violates Armijo, it contracts by halving until Armijo holds or a safeguard floor is reached. This makes the optimizer robust to mis-scaled step_size inputs (including the Newton-natural α=1 that BFGS expects on well-conditioned quadratics) without forcing the caller to retune it, and preserves the secant pair (s, y) curvature condition so the L-BFGS inverse-Hessian approximation stays SPD.

All inner products use the manifold metric g_x(·,·), and every secant pair is parallel-transported into the current tangent space before it enters the two-loop recursion, so the BFGS algebra never mixes vectors living in different tangent spaces (the bug fixed in #616). The freshly formed secant pair likewise transports the accepted step from the old tangent space into the new one before pairing it with the gradient difference, so both s and y live in T_{x_new}M.

Trait Implementations

impl Clone for RiemannianLBFGS

fn clone(&self) -> RiemannianLBFGS

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for RiemannianLBFGS

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for RiemannianLBFGS

fn default() -> Self

Returns the “default value” for a type.

impl PartialEq for RiemannianLBFGS

fn eq(&self, other: &RiemannianLBFGS) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for RiemannianLBFGS