Module krylov 

Krylov subspace methods for sparse eigenvalue computation.

This module provides GPU-accelerated Lanczos and Arnoldi iteration for computing extreme eigenvalues and eigenvectors of large sparse matrices.

• LanczosPlan – Lanczos iteration for symmetric matrices, producing a tridiagonal matrix whose eigenvalues approximate those of the original matrix.
• ArnoldiPlan – Arnoldi iteration for general (non-symmetric) matrices, producing an upper Hessenberg matrix.

Both methods rely on repeated SpMV (sparse matrix-vector multiplication) as the core computational primitive, combined with vector orthogonalization kernels generated as PTX at runtime.

Structs

ArnoldiConfig

Configuration for Arnoldi iteration on general sparse matrices.

ArnoldiPlan

Arnoldi iteration plan for general sparse eigenvalue problems.

ArnoldiResult

Result of an Arnoldi iteration.

LanczosConfig

Configuration for Lanczos iteration on symmetric sparse matrices.

LanczosPlan

Lanczos iteration plan for symmetric sparse eigenvalue problems.

LanczosResult

Result of a Lanczos iteration.

Enums

EigenTarget

Specifies which eigenvalues to target in Krylov iteration.

Constants

KRYLOV_BLOCK_SIZE

Default block size for Krylov vector operations.