Dataset from: "Enabling Structure-Only Initialization and Out-of-Distribution Generalization in GNN-based Molecular Dynamics Simulators"

Overview This repository contains molecular dynamics simulation datasets capturing the uniaxial compression of disordered elastic networks (DENs). The networks were generated and simulated using LAMMPS. This dataset is designed for training, validation, and benchmarking of Graph Neural Network-based MD simulators and structural optimization pipelines for mechanical metamaterials. The dataset spans a range of networks with different topologies, exhibiting a wide range of dynamics and Poisson's ratios. The dataset includes two distinct types of optimized auxetic elastic networks: Global Node Displacement Optimization: Networks optimized by tuning local node coordinates. Stiffness-Based Optimization: Networks optimized by tuning individual harmonic bond stiffnesses. For full methodological details, please refer to the main manuscript and Supplementary Information: [arXiv:2605.09495]. Dataset Structure & Contents We provide two variants of the dataset : data.tar.gz (Full Version) data_mini.tar.gz (Mini Version) Dataset is split into chunks, where each chunk contains a single unique DEN compression simulation.概述 本仓库收录了用于模拟无序弹性网络（disordered elastic networks, DENs）单轴压缩过程的分子动力学（Molecular Dynamics, MD）仿真数据集。所有网络均通过LAMMPS软件生成并完成仿真。本数据集可为基于图神经网络（Graph Neural Network, GNN）的MD仿真器，以及面向力学超材料的结构优化流程，提供训练、验证与基准测试支持。数据集涵盖了多种不同拓扑结构的网络，可展现出宽泛的动力学行为与泊松比分布。 本数据集包含两类经优化的拉胀弹性网络： - 全局节点位移优化：通过调整局部节点坐标实现网络优化 - 基于刚度的优化：通过单独调控各谐和键的刚度实现网络优化 如需获取完整的方法学细节，请参阅主论文与补充材料：[arXiv:2605.09495] 数据集结构与内容 我们提供两种数据集变体： - data.tar.gz（完整版） - data_mini.tar.gz（迷你版） 数据集被划分为多个数据块，每个数据块包含一组独立的无序弹性网络单轴压缩仿真任务。