Behind the Matrix: Predicting Numerical Error in Generalized Eigenvalue Solvers

Overview

This repository contains the code and data used in the paper "Behind the Matrix: Predicting Numerical Error in Generalized Eigenvalue Solvers" by Anna Diaz Farias.

Repository Contents

• matrix_generator.py — Python script to generate synthetic SPD matrix pairs and compute relative forward error. Produces matrix_dataset.csv.
• matrix_dataset.csv — Full factorial dataset of 600 synthetic SPD matrix pairs with computed relative forward error across combinations of dimension n, scaling factor s, and condition number κ.
• CC_proj.R — R script for all statistical modeling, stepwise selection, and figure generation.

Requirements

Python

• Python 3.9+
• NumPy
• SciPy
• pandas

R

• R 4.2+
• tidyverse
• ggplot2
• scales
• emmeans

Usage

1. Run matrix_generator.py to generate the dataset
2. Run CC_proj.R to reproduce all models and figures

Contact

For questions, contact Anna Diaz at anna.diaz0702@gmail.com