Struct IlukFactorization 

pub struct IlukFactorization<T: GpuFloat> {
    pub lu: CsrMatrix<T>,
    pub diag_inv: Vec<T>,
    pub fill_level: usize,
}

ILU(k) factorization result stored as combined L\U in CSR.

L is unit lower triangular (diagonal = 1, stored implicitly) and U is upper triangular (diagonal stored explicitly). Both share the CSR storage lu.

Fields

lu: CsrMatrix<T>

Combined L\U factors in CSR format. Lower triangle contains L (without unit diagonal). Upper triangle (including diagonal) contains U.

diag_inv: Vec<T>

Inverse of the diagonal of U, for efficient forward/backward solve.

fill_level: usize

The fill level used to compute this factorization.

Implementations

impl<T: GpuFloat> IlukFactorization<T>

pub fn compute( _handle: &SparseHandle, matrix: &CsrMatrix<T>, config: &IlukConfig, ) -> SparseResult<Self>

Compute ILU(k) from a CSR matrix.

Arguments

• _handle – Sparse handle (reserved for GPU path).
• matrix – Square CSR matrix to factor.
• config – ILU(k) configuration (fill level).

Errors

Returns SparseError::DimensionMismatch if the matrix is not square. Returns SparseError::SingularMatrix if a zero pivot is encountered.

pub fn apply(&self, r: &[T], z: &mut [T]) -> SparseResult<()>

Apply as preconditioner: solve (L*U)*z = r.

Forward solve L*y = r then backward solve U*z = y.

Arguments

• r – Right-hand side vector (host, length n).
• z – Output vector (host, length n).

Errors

Returns SparseError::DimensionMismatch if vector lengths are wrong.