An agentic AI framework that ran atomistic simulations and found previously unobserved dislocations in nickel grain boundaries
AspireAI community · arXiv
Agentic AI systems hold potential for automating complex workflows in materials discovery and design, but extending them to domain-intensive mechanistic studies remains challenging. General grain boundaries exemplify that challenge: structurally complex, common in real polycrystals, and still understudied. GBClaw is an agentic AI framework for investigating the structure and deformation of general grain boundaries, integrating large language models into a modular architecture that combines systematic simulations with targeted mechanistic analysis through iterative workflow refinement and dynamic replanning. Applied to 14 general asymmetric tilt grain boundaries in nickel, the framework characterised interfacial defect structures and deformation modes across these incommensurate boundaries, establishing a general approach to identify previously unobserved, widely spaced secondary grain-boundary misfit dislocations alongside the closely spaced primary ones in both simulated atomic configurations and experimental images. Comparisons of grain-boundary sliding with and without the secondary dislocations revealed substantial differences in shear strength, underscoring the critical role of misfit-dislocation hierarchies in sliding resistance, and researcher-guided follow-up calculations elucidated the structural origins of shear asymmetry and an anomalous shear response. The limitation is that this is a demonstration in one materials system, and whether the architecture transfers to domains where the agent cannot generate its own experimental data is untested here.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.