From Nanoscale Interface to Macroscopic Conductivity: A Physics-Guided Graph Framework for Conductive Polymer Composites

The rational design of conductive polymer nanocomposites is hindered by an unresolved challenge: establishing a quantitative, predictive, and mechanistically interpretable link between nanoscale filler–polymer interactions and macroscopic electronic performance. Here, we present a physics-informed graph learning framework that integrates hybrid particle-field molecular dynamics with graph neural networks to decode this structure–property relationship in carbon nanotube (CNT)/triblock copolymer polystyrene-butadiene-styrene (SBS) systems. Through a physics-guided feature screening strategy, we identify the CNT/polystyrene (PS) interface as dominant structural descriptor for electrical conductivity, effectively filtering out polybutadiene (PB)-phase descriptors that are not directly relevant to conductivity. To overcome the limitations of purely data-driven models, we embed the classical percolation power law into a physics-informed graph convolutional network (PIGCN), enabling accurate predictions even in data-sparse regions near the percolation threshold at unseen concentrations. The convergence dynamics of key parameters (the prefactor η, percolation threshold fc, and exponent t) provide dynamic insights into network formation. Experimental validation on SBS/CNT composites with controlled aspect ratios confirms predictions: high-aspect-ratio CNTs establish long-range connectivity, whereas low-aspect-ratio fillers enhance network robustness. From these insights, we derive actionable design principles, including targeting the CNT/PS interface, leveraging aspect-ratio complementarity, and defining critical loading windows, establishing a knowledge-driven paradigm that bridges physics-constrained learning with experiment for the accelerated rational design of high-performance nanocomposites.