Module matrix_powers 

Sparse matrix powers and polynomial evaluation.

Computes A^k via repeated SpGEMM with binary exponentiation and optional structure reuse optimization. Also provides matrix polynomial evaluation via Horner’s method: p(A) = c0*I + c1*A + c2*A^2 + ... + ck*A^k.

All operations work on CSR (Compressed Sparse Row) format stored as (row_offsets, col_indices, values) triple on the host side.

(C) 2026 COOLJAPAN OU (Team KitaSan)

Structs

MatrixPowerConfig

Configuration for sparse matrix power computation.

MatrixPowerResult

Result of a sparse matrix power computation.

Functions

add_csr

Add two CSR matrices with the same number of rows: C = A + B.

estimate_power_nnz

Heuristic estimate of nnz(A^k) based on average degree.

host_spgemm

Host-side sparse matrix-matrix multiplication: C = A * B.

scalar_multiply_csr

Multiply all values of a CSR matrix by a scalar, keeping the sparsity structure unchanged.

sparse_identity

Construct the CSR representation of an n x n identity matrix.

sparse_matrix_polynomial

Evaluate a polynomial p(A) = c0*I + c1*A + c2*A^2 + ... + ck*A^k using Horner’s method.

sparse_matrix_power

Compute A^power for a sparse CSR matrix using binary exponentiation.