BlockSparseMatrices.jl
BlockSparseMatrices.jl provides a representation for sparse matrices that are composed of a limited number of (dense) blocks. It also includes specialized algorithms for symmetric block‑sparse matrices, storing only the necessary half of the off‑diagonal blocks.
Key Features
- Block‑sparse storage – the matrix is built from a small set of (dense) sub‑blocks.
- Symmetric support – for symmetric block‑sparse matrices only the lower (or upper) triangular block‑structure is kept, reducing memory usage.
- Multithreaded matrix–vector multiplication – leverages OhMyThreads and GraphsColoring for safe parallelism.
Implemented Operations
| Operation | Description |
|---|---|
* (matrix‑vector product) | Fast, multithreaded multiplication. |
transpose / adjoint | Returns the (adjoint) transpose of a block‑sparse matrix. |
getindex | Access individual elements of the matrix. |
setindex! | Modify elements in‑place. |
| Visualization | Visual inspection via UnicodePlots. |
sparse | Convert to a standard SparseMatrixCSC from SparseArrays. |
Related Packages
If you need alternative block‑matrix representations or additional functionality, consider: