Geometric Encoding of Thermal History in Glasses: Strain Topology as a Learnable Structural Signature

Geometric Encoding of Thermal History in Glasses: Strain Topology as a Learnable Structural Signature Abstract / Description: The glass transition poses a fundamental question in condensed matter physics: how does thermal history become encoded in the atomic structure of disordered solids? Here, we demonstrate that the memory of cooling rate is geometrically encoded in the strain topology—the spatial distribution of local mechanical distortions—rather than in simple density or coordination metrics. Using graph neural networks (GNNs) as structural probes and a systematic ablation methodology, this repository contains the data, models, and analysis demonstrating this encoding across multiple distinct glass-forming systems. Key Findings & Updates in this Version: Binary Classification on Lennard-Jones (LJ): We achieve 95.10% ± 1.02% classification accuracy between fast-cooled and slow-cooled LJ glasses under 5-fold stratified cross-validation. Accuracy remains robust (90-94%) even when feature magnitudes are instance-normalized to remove trivial separability, whereas pure topological features fail at near-random accuracy (52%). Cross-System Validation (Kob-Andersen Mixture): To prove this is not an artifact of the LJ potential, we applied the identical GATv2 framework to the Kob-Andersen binary mixture ($N=4096$) from the independently published GlassBench dataset. The network achieved 96.23% ± 0.74% accuracy using only pure geometric (5D) features, requiring no local Hessian input Continuous Encoding & Structural Fictive Temperature: We extend the binary framework to a continuous thermodynamic variable by generating glasses across eight logarithmically spaced cooling rates. Principal Component Analysis of the GNN latent space reveals a single dominant axis (PC1) that captures 76.9% of all structural variance. This axis acts as a purely geometric proxy for fictive temperature and is approximately 73x more sensitive to cooling rate than potential energy per particle. Revised Memory Horizon Transition: We show that GNNs can partially recover ordered initial geometries from thermally scrambled glasses with ~92.3% shape fidelity. High-resolution mapping of this reconstruction fidelity reveals an ultra-sharp transition: geometric memory collapses abruptly within the first 50 scrambling steps, which is substantially earlier than previously estimated. Physical Manifestation: Visualization of strain-weighted contact networks confirms the physical picture: fast-cooled glasses exhibit fragmented stress patterns, while slow-cooled glasses develop extended, percolating force chains. These results establish that thermal history manifests as a learnable, reproducible geometric pattern across qualitatively different glass-forming systems.