Skip to main content

svds

Function svds 

Source
pub fn svds<T, O>(
    op: &mut O,
    k: usize,
    opts: &SvdsOptions<T>,
) -> Result<Svd<T>, StatlaError>
where T: Real, O: AdjointOperator<T> + ?Sized,
Expand description

Leading k singular triplets by Golub–Kahan–Lanczos bidiagonalization with thick restart — a Krylov engine of its own, not an eigensolver on AᵀA (S2): it never squares the condition, so every singular value comes out with an absolute error of order ε σ_1.

Algorithm B of Chapter 12.5: full two-sided DGKS reorthogonalization, one-sided Jacobi SVD of the projected matrix, thick restart in the arrowhead form of Baglama & Reichel, locking of converged triplets, absolute convergence test |ρ_i| ≤ τ σ̃_1.

Guarantee. Every returned triplet has ‖(A v − σ̂ u; Aᵀ u − σ̂ v)‖ ≤ (τ + c_G·ncv·u) σ̂_1 (c_G = 8, measured; Chapter 12.5), hence some singular value σ_j (or 0) lies within ‖r‖/√2 of σ̂ (Jordan–Wielandt; proved in the chapter). The index j is the i-th unless a copy of a multiple value was missed.

§Errors

StatlaError::InvalidArgument (k = 0; k ≥ min(m, n); ncv ∉ (k, min(m, n)]; tol < 0 or not finite; reorth_eta ∉ (0, 1); v0 = 0), StatlaError::Dimension (v0), StatlaError::NonFinite (v0, or the operator returned NaN/∞), StatlaError::Operator, StatlaError::NoConvergence.