Data and code for: Energy, exit barriers, and tree persistence as predictors of escape times in small spherical p-spin glasses: a trajectory-level test

Data and code accompanying the manuscript "Energy, exit barriers, and tree persistence as predictors of escape times in small spherical p-spin glasses: a trajectory-level test".

Contents

• analysis/new_experiment.py — full production pipeline: tangent-space gradient descent + Newton polishing for minimum sampling, string-method minimax barrier calculation, and projected-noise Langevin dynamics with destination auditing.
• analysis/new_analysis.py — statistical analysis and figure generation (Harrell's C, Cox regression, Kaplan–Meier survival curves, Arrhenius fits).
• experiment/ — raw and processed results: 142 sampled minima (N = 10–18), 624 string-method barriers, 4544 escape trajectories, plus summary statistics and run_summary.json.
• figures_new/ — the four figures used in the manuscript.
• checksums.md5 — file integrity manifest.

Reproducibility: all randomness is seeded; rerunning the two pipeline scripts regenerates every number and figure reported in the paper. The production dynamics use t_max = 300 (see run_summary.json).

The code runs on Python 3.13 with NumPy/SciPy/Matplotlib only.