Reproduce “PaddleMaterials: An End-to-End Deep Learning Toolkit for Materials Science” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xuwei Liu, Zhimin Zhang, Xiaoguang Hu, Zhongwei Liu, Yuanhui Sun, Jun Huo, Hua Wu, Haifeng Wang
PaddleMaterials connects materials workflows for generation, relaxation, and property prediction while preserving task-specific data representations and scientific conventions. Its shared architecture organizes models, data, and calculations without imposing one representation on every task. In an eight-atom Li₂Mn₂O₄ example, formation-energy predictions on generated and relaxed geometries were within 7 meV per atom of same-geometry PBE+U references; this does not establish material viability or general accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.