---
license: apache-2.0
language:
- en
tags:
- linear-systems
- sparse-matrices
- scientific-computing
- magnetic-diffusion
- benchmark
pretty_name: LinearSolveBench magnetic diffusion development data
size_categories:
- n<1K
---

# LinearSolveBench: magnetic diffusion development data

The 96 development cases for the `magnetic_diffusion_flash` family of [LinearSolveBench](https://www.autodidakt.ai/linear-solve-bench).\
Benchmark code, family manifests, the reference solver, and evaluation rules live in the [code repository](https://github.com/hgarud/LinearSolveBench).

## Dataset

These scalar magnetic-diffusion linear systems were captured from AlWire and two-dimensional driven ZPinch test-problems from FLASH 4.8.\
These captures cover startup time steps and support independent linear solves, not complete
simulation trajectories.\
See the [FLASH diffusion documentation](https://flash.rochester.edu/site/flashcode/user_support/flash_ug_devel/node128.html)
for the underlying formulation.

Matrices are real `float64` CSR, with 9,984–113,664 rows and
108,364–1,245,184 nonzeros.

## Files

- `cases/`: one ZIP archive per development case.
  dimensions, and source-run provenance.
- `README.md` and `LICENSE`: dataset documentation and license.

Each `cases/flash-replay-NNNN.zip` contains exactly:

- `matrix.npz`: coefficient matrix `A` in SciPy CSR format.
- `b.npy`: captured right-hand side `b`.
- `x0.npy`: captured initial guess `x0`.
- `case.json`: tolerances, provenance identifiers, and file and numerical-array
  checksums.