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 2303.04791cs.LGcond-mat.mtrl-sciphysics.chem-phphysics.comp-ph
收藏

用于分子图的基于 Ewald 的长程消息传递

Ewald-based Long-Range Message Passing for Molecular Graphs

Max Planck Institute for the Science of Light(马克斯·普朗克光科学研究所);Friedrich-Alexander-Universität Erlangen-Nürnberg(埃尔朗根-纽伦堡弗里德里希-亚历山大大学);Ludwig-Maximilians-Universität München(慕尼黑路德维希-马克西米利安大学);Technical University of Munich(慕尼黑工业大学);Google Research(谷歌研究院)

Arthur Kosmala, Johannes Gasteiger, Nicholas Gao, Stephan Günnemann

AI总结:

本文提出基于 Ewald 求和的 Fourier 空间非局部消息传递方法,以弥补 MPNN 距离截断难以学习长程相互作用的不足,并在 OC20 与 OE62 上跨四个模型显著降低能量误差。

摘要:

近年来,从分子数据中学习势能面的神经架构取得了快速进展。这一成功的关键驱动力之一是 Message Passing Neural Network(MPNN,消息传递神经网络)范式。它随系统规模良好扩展的能力,部分依赖于对消息施加空间距离限制。虽然这种对局部性的关注是一种有用的归纳偏置,但它也阻碍了对静电作用和 van der Waals 力等长程相互作用的学习。为解决这一缺陷,我们提出 Ewald 消息传递:这是一种非局部的 Fourier 空间方案,通过对频率而非距离设置截断来限制相互作用,并在理论上以 Ewald 求和方法为坚实基础。由于其计算成本低且与架构细节无关,它可以作为现有 MPNN 架构之上的增强模块。我们使用四个基线模型和两个包含多样周期性结构(OC20)与非周期性结构(OE62)的数据集测试了该方法。我们观察到,在所有模型和数据集上,能量平均绝对误差均获得稳健改善,在 OC20 上平均降低 10%,在 OE62 上平均降低 16%。分析表明,对于真实能量中长程贡献较高的结构,这些改善的影响尤为突出。

英文摘要:

Neural architectures that learn potential energy surfaces from molecular data have undergone fast improvement in recent years. A key driver of this success is the Message Passing Neural Network (MPNN) paradigm. Its favorable scaling with system size partly relies upon a spatial distance limit on messages. While this focus on locality is a useful inductive bias, it also impedes the learning of long-range interactions such as electrostatics and van der Waals forces. To address this drawback, we propose Ewald message passing: a nonlocal Fourier space scheme which limits interactions via a cutoff on frequency instead of distance, and is theoretically well-founded in the Ewald summation method. It can serve as an augmentation on top of existing MPNN architectures as it is computationally inexpensive and agnostic to architectural details. We test the approach with four baseline models and two datasets containing diverse periodic (OC20) and aperiodic structures (OE62). We observe robust improvements in energy mean absolute errors across all models and datasets, averaging 10% on OC20 and 16% on OE62. Our analysis shows an outsize impact of these improvements on structures with high long-range contributions to the ground truth energy.

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。

↑