Reproduce “A cyclical route linking fundamental mechanism and AI algorithm: An example from tuning Poisson's ratio in amorphous networks” on OpenResearch

About this paper

Changliang Zhu, Chenchao Fang, Zhipeng Jin, Baowen Li, Xiangying Shen, Lei Xu

This research outlines a cyclical "AI for Science" route, showing how artificial intelligence aids in discovering the fundamental mechanism governing Poisson's ratio in amorphous networks, which subsequently improves AI algorithms for material property prediction. A simulated annealing algorithm tuned Poisson's ratio across -0.9 to 0.9, and physically-informed deep learning models achieved 96 times faster training and 25 times faster prediction with an R^2 of 0.98 and RMSE of 0.02.

Key insights

• The Poisson's ratio in amorphous networks is predominantly determined by the weighted superposition of a few low-frequency vibrational normal modes of the dynamical matrix, a mechanism universally applicable across different amorphous systems.
• The tuning of Poisson's ratio is explained as a competition between the contributions of distinct normal modes having positive and negative effective Poisson's ratios, with their weights shifting during optimization.
• Integrating fundamental physical insights (e.g., using the dynamical matrix as input) into AI algorithms significantly enhances their efficiency, accuracy, and interpretability in predicting material properties, compared to generic visual inputs.

Summary by alphaXiv.