pub struct SymShiftInvert<T: Real = f64, K: DenseKernel<T> = NaiveDense> {
sigma: T,
n: usize,
below: usize,
ldl: SupernodalLdl<T, K>,
num: NumericLdl<T>,
low: CscMatrix<T>,
r: Vec<T>,
}Expand description
x ↦ (A − σI)⁻¹ x for a symmetric sparse A (lower triangle read), by
SupernodalLdl<T, K> with no regularization; the inertia of A − σI
is kept: by Sylvester’s law it counts the eigenvalues below σ
(SymShiftInvert::count_below) — a check ARPACK does not offer. Each
application refines once (Algorithm R: two solves and one product with
the kept lower triangle of A − σI). The operator of eigsh_near
(Algorithm SI of Chapter 12.6.2).
Fields§
§sigma: T§n: usize§below: usize§ldl: SupernodalLdl<T, K>§num: NumericLdl<T>§low: CscMatrix<T>The lower triangle of M = A − σI (Algorithm R).
r: Vec<T>The residual r of step R2.
Implementations§
Source§impl<T: Real, K: DenseKernel<T>> SymShiftInvert<T, K>
impl<T: Real, K: DenseKernel<T>> SymShiftInvert<T, K>
Sourcepub fn new(a: &CscMatrix<T>, sigma: T, kernel: K) -> Result<Self, StatlaError>
pub fn new(a: &CscMatrix<T>, sigma: T, kernel: K) -> Result<Self, StatlaError>
Analyze (own AMD ordering) and factorize A − σI.
§Errors
StatlaError::InvalidArgument (not square), StatlaError::NonFinite
(a, sigma), StatlaError::Kkt (a structurally invalid a; a
zero pivot: σ is numerically an eigenvalue).
Sourcepub fn count_below(&self) -> usize
pub fn count_below(&self) -> usize
Number of eigenvalues of A strictly below σ (negative inertia of
A − σI).
Trait Implementations§
Source§impl<T: Debug + Real, K: Debug + DenseKernel<T>> Debug for SymShiftInvert<T, K>
impl<T: Debug + Real, K: Debug + DenseKernel<T>> Debug for SymShiftInvert<T, K>
Source§impl<T: Real, K: DenseKernel<T>> LinearOperator<T> for SymShiftInvert<T, K>
impl<T: Real, K: DenseKernel<T>> LinearOperator<T> for SymShiftInvert<T, K>
Source§fn apply_block(
&mut self,
x: MatRef<'_, T>,
y: MatMut<'_, T>,
) -> Result<(), OperatorError>
fn apply_block( &mut self, x: MatRef<'_, T>, y: MatMut<'_, T>, ) -> Result<(), OperatorError>
Y ← A X for a block of columns; the default loops over
LinearOperator::apply. Dense and sparse implementations
override it (one pass over A for l columns — the cost that
matters in crate::rsvd()). Read more