Machine learning-driven atomistic simulation of mechanical deformation in nanoporous amorphous carbon

Abstract

Nanoporous amorphous carbon (NP α-C) is a promising material for next-generation energy storage systems, particularly as a key component in lithium-ion battery anodes. However, its disordered atomic structure and complex nanoscale porosity pose significant challenges for understanding its structure-property relationships. In this study, we generated and analyzed over 200,000 unique NP α-C configurations using the Gaussian Random Field Method combined with machine learning-driven molecular dynamics simulations. This approach enabled the creation of an extensive structural database, covering porosities from 10 % to 90 %, average pore sizes from 5 to 60 Å, and pore size variances from 0 to 60 Å². Our findings reveal that pore structure plays a crucial role in governing the elastic and plastic behavior of NP α-C. Under triaxial tension, stress concentrates at ligament-junction regions, leading to ligament thinning, single-chain formation, and eventual fracture. Cyclic loading tests further demonstrate that most fractures occur in the first cycle, with minimal crack propagation and a significant reduction in elastic constants in subsequent cycles. This study establishes a robust theoretical framework for optimizing NP α-C microstructures, offering valuable insights into the design of high-performance porous materials for energy storage applications.