arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新，休息一下吧，周末愉快~~
arXiv 2605.08960cond-mat.mtrl-scics.LGphysics.chem-phphysics.comp-ph
收藏

CrystalREPA: 从通用机器学习 interatomic 势能中转移物理先验到晶体生成模型

CrystalREPA: Transferring Physical Priors from Universal MLIPs to Crystal Generative Models

AI for Science Institute, Beijing, China（人工智能科学研究院，北京，中国）；Center for Data Science, Peking University, Beijing, China（数据科学中心，北京大学，北京，中国）；LMAM and School of Mathematical Sciences, Peking University, Beijing, China（LMAM与数学科学学院，北京大学，北京，中国）；Center for Machine Learning Research, Peking University, Beijing, China（机器学习研究中心，北京大学，北京，中国）；National Key Laboratory of Computational Physics, Institute of Applied Physics and Computational Mathematics, Beijing, China（国家计算物理重点实验室，应用物理与计算数学研究所，北京，中国）；HEDPS, CAPT, College of Engineering, Peking University, Beijing, China（HEDPS、CAPT、工程学院，北京大学，北京，中国）

Chengqian Zhang, Yucheng Jin, Duo Zhang, Tiejun Li, Han Wang

备注：
NeurIPS 2026 (Spotlight)
阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

CrystalREPA 通过在训练时对齐生成编码器的原子隐藏状态与冻结的 MLIP 表示，转移稳定性感知的原子先验，提升生成晶体的热力学稳定性、结构有效性及结构保真度。

摘要：

晶体生成模型主要学习稳定晶体的外观，缺乏对稳定性的显式监督。通过能量探测，我们揭示了最先进的晶体生成模型与预训练的通用机器学习 interatomic 势能（MLIPs）之间存在显著的表示差距，并展示这种差距可通过简单的训练时对齐来关闭。我们提出了 Crystal REPresentation Alignment（CrystalREPA），一种即插即用的框架，通过元素感知的对比目标对生成编码器的原子隐藏状态与冻结的 MLIP 表示进行对齐，以转移具有稳定性的原子先验，仅需边际训练开销且无额外推断成本。在三种生成框架、十个 MLIP 教师和两个基准数据集上，CrystalREPA 一致提升了生成晶体的热力学稳定性、结构有效性及结构保真度。同样重要的是，我们发现 MLIP 的转移效果难以通过其在标准排行榜上的准确性（如 Matbench Discovery）来预测，但能通过其原子表示空间的区分性来预测，从而提供了一个实用且不依赖准确性的标准，用于选择用于生成转移的 MLIP 教师。

英文摘要：

Crystal generative models mainly learn what stable crystals look like, with little explicit supervision for what makes them stable. We reveal a substantial representation gap between state-of-the-art crystal generative models and pretrained universal machine learning interatomic potentials (MLIPs) via energy probing, and show this gap can be closed by a simple training-time alignment. We propose Crystal REPresentation Alignment (CrystalREPA), a plug-and-play framework that aligns the atom-wise hidden states of generative encoders with frozen MLIP representations through an element-aware contrastive objective, transferring stability-aware atomistic priors with marginal training overhead and no additional inference cost. Across three generative frameworks, ten MLIP teachers, and two benchmark datasets, CrystalREPA consistently improves the thermodynamic stability, structural validity, and structural fidelity of generated crystals. Equally important, we find that an MLIP's transfer effectiveness is poorly predicted by its accuracy on standard leaderboards (e.g., Matbench Discovery) but strongly predicted by the distinguishability of its atom-wise representation space, yielding a practical, accuracy-independent criterion for selecting MLIP teachers for generative transfer. Code is available at https://github.com/Chengqian-Zhang/CrystalREPA.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑