solver-advisor 0.1.0

Solver Advisor AI‑powered matrix diagnostics and solver recommendations for scientific computing

Solver Advisor is a lightweight analysis tool designed to inspect sparse matrices from scientific simulations (FEM, CFD, optimization, power systems, structural mechanics, etc.) and automatically recommend suitable iterative solvers and preconditioners.

It analyzes matrix structure, symmetry, SPD‑properties, block patterns, condition number, and spectral characteristics using Lanczos iterations. The tool provides actionable solver recommendations such as CG, GMRES, MINRES, and preconditioners like Jacobi, ILU, SSOR, or Multigrid.

Features Matrix loading (.mtx, .csv, .npy, .npz)

Symmetry detection

SPD (symmetric positive definite) test

Block‑structure detection

Condition number estimation

Largest/smallest eigenvalue estimation

Lanczos spectrum approximation

Solver recommendation (CG, GMRES, MINRES)

Preconditioner recommendation (Jacobi, SSOR, ILU, Multigrid)

CLI interface

GUI interface (Tkinter)

SuiteSparse matrix downloader (Python script)

Project Structure Kood solver-advisor/ │ ├── solver_advisor/ │ ├── io.py # Matrix loading utilities │ ├── analysis.py # Symmetry, SPD, block detection │ ├── spectrum.py # Eigenvalue & Lanczos routines │ ├── diagnostics.py # Solver & preconditioner logic │ ├── run.py # High-level execution wrapper │ ├── gui/ │ ├── app.py # Tkinter GUI │ ├── cli/ │ ├── main.py # CLI entry point │ ├── matrices/ # Downloaded matrices │ ├── tools/ │ ├── download_matrices.py # SuiteSparse downloader │ ├── tests/ │ ├── test_symmetry.py │ ├── test_spd.py │ └── README.md Installation Clone the repository:

Kood git clone https://github.com//solver-advisor cd solver-advisor Install in editable mode:

Kood pip install -e . This makes the solver-advisor CLI command available system‑wide.

Usage CLI Analyze a matrix directly from the terminal:

Kood solver-advisor matrices/example.mtx Example output:

Kood Form: (5000, 5000) Symmetry: True SPD: True Block structure: False Condition number: 1.2e7 Solver: CG Preconditioner: ILU/Multigrid GUI Start the graphical interface:

Kood python gui/app.py The GUI allows you to:

Select matrices from the matrices/ folder

Run full diagnostics

View solver recommendations

Inspect eigenvalues, condition number, SPD status, block structure, etc.

Downloading Matrices (SuiteSparse) Use the provided script to download real-world sparse matrices:

Kood python tools/download_matrices.py This downloads and extracts matrices from the SuiteSparse Matrix Collection into the matrices/ directory.

Example Matrix You can generate a small example matrix:

python from solver_advisor.io import create_example_matrix create_example_matrix() This creates matrices/example.mtx.

Requirements Python 3.10+

NumPy

SciPy

ssgetpy

Tkinter (included with most Python installations)

License MIT License (or whichever you choose)

Author Allar‑Joel Möldre
Numerical Analysis • HPC • Solver Diagnostics

Future Work AMG preconditioner integration

PETSc backend support

Web‑based GUI

Matrix pattern visualization

Solver performance prediction

Automatic preconditioner tuning