标题： 聚合物积分方程理论的机器学习闭合方法 显示英文标题

标题： A Machine Learning Closure for Polymer Integral Equation Theory

摘要： 聚合物参考相互作用位点模型（PRISM）理论，是奥恩斯坦-泽尔尼克液体状态理论的后代，是一种强大的工具，用于预测平衡聚合物系统的结构和热力学性质，但在某些重要情况下，其准确性和适用性可能受到限制。通常，这些缺点可归因于用于求解积分方程的解析闭合关系。在这里，我们提出了一种基于机器学习（ML）的闭合关系，该关系是通过对均聚物熔体和溶液的粗粒度分子动力学模拟数据集进行训练得到的。使用ML闭合关系的PRISM理论在预测典型粗粒度模型系统的结构方面优于传统的原子闭合关系（例如，Percus-Yevick）。我们还使用ML闭合关系准确地模拟了小角中子散射实验的结果。因此，这种增强的ML PRISM理论可以促进快速软材料的发现和设计工作。 显示英文摘要

摘要： Polymer reference interaction site model (PRISM) theory, a descendent of Ornstein-Zernike liquid state theory, is a powerful tool to predict the structure and thermodynamics of equilibrium polymer systems, but its accuracy and applicability can be limited in some important cases. Typically, these shortcomings are traced to the analytical closure relationships used to solve the integral equations. Here, we propose a machine learning (ML)-based closure relation trained on a dataset of coarse-grained molecular dynamics simulations of homopolymer melts and solutions. PRISM theory with the ML closure outperforms traditional atomic closures (e.g., Percus-Yevick) in predicting the structure of typical coarse-grained model systems. We also use the ML closure to accurately model the results of small-angle neutron scattering experiments. This ML-enhanced PRISM theory can therefore enable rapid soft materials discovery and design efforts.