Struct Projection 

pub struct Projection { /* private fields */ }

The L2 (Galerkin) projection from src onto dst, kept factored as G⁻¹ M with G the (banded) Gram matrix of dst and M = ⟨dst, src⟩; G⁻¹ M itself is generally dense.

Implementations

impl Projection

pub fn new(src: &Space, dst: &Space, weight: Weight) -> Result<Self>

pub fn n_src(&self) -> usize

pub fn n_dst(&self) -> usize

pub fn cross(&self) -> &Csr

M = ⟨dst_i, src_j⟩_w.

pub fn apply(&self, f: ArrayView1<'_, f64>) -> Array1<f64>

pub fn apply_last_axis(&self, f: ArrayView2<'_, f64>) -> Array2<f64>

Projects each row of f, shape (k, n_src), giving (k, n_dst).

pub fn apply_bounded_last_axis( &self, f: ArrayView2<'_, f64>, bounds: Bounds, ) -> Array2<f64>

The best bounded approximation: the coefficients c minimizing the weighted L2 error ‖f − Σ c_i dst_i‖ subject to the bounds and to conservation, Σ c_i ∫ dst_i w = Σ g_i ∫ dst_i w for the unbounded projection g. Equivalently, c minimizes (c − g)ᵀ G (c − g), a quadratic program with a banded Hessian, solved by a primal-dual active-set method. Because the error is measured on the function, corrections stay local: clipped mass is put back next to where it was removed.

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

The dense matrix G⁻¹ M, shape (n_dst, n_src).

Trait Implementations

impl Clone for Projection

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Projection

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

Formats the value using the given formatter.