开发物理信息神经框架MEOWN，用于快速预测晶体材料中μ子停止位点，以理解采用μ子谱学的量子磁体

中文摘要 AI 辅助

μ子自旋旋转/弛豫是理解新兴量子材料中磁序、自旋动力学、超导性和复杂磁相的最强大微观探针之一。定量解释关键依赖于对μ子停止位点的准确识别，这一问题传统上通过基于密度泛函理论的结构弛豫来解决。然而，传统DFT方法计算要求高、耗时，且需要大量计算。在此，我们开发了MEOWN（用于优化加权网络的μ子引擎），这是一个基于机器学习、物理信息的计算框架，结合了可极化无扰静电势（P-UEP）模型与机器学习引导的优化和对称性驱动的弛豫，以预测晶体固体中能量上有利的μ子停止位点。MEOWN明确将静电相互作用、电子屏蔽、极化效应和零点运动纳入学习流程，提供物理上可解释的预测，同时大幅降低计算成本，并保持物理一致性和科学严谨性。为验证该框架，我们研究了几个成熟基准材料，包括MnSi、CoF2、CaF2、LiF和NaF。预测的μ子停止位点和偶极场与先前报道的DFT+μ结果密切一致，证明了该方法在化学多样性系统中的可靠性和可转移性。值得注意的是，MEOWN在几分钟内预测平衡μ子停止位点，相比传统基于DFT的计算提供了显著加速。本手稿介绍了MEOWN作为通用软件的理论基础、计算方法和验证，用于晶体材料中快速可靠的μ子停止位点预测。

英文摘要

Muon-spin rotation/relaxation is one of the most powerful microscopic probes for understanding magnetic order, spin dynamics, superconductivity, and complex magnetic phases in emerging quantum materials. Quantitative interpretation depends critically on the accurate identification of the muon stopping site, a problem traditionally addressed using density functional theory-based structural relaxation. However, conventional DFT approaches are computationally demanding, time-consuming, and require extensive calculations. Here, we develop MEOWN (Muon Engine for Optimized Weighted Networks), a machine-learning-based, physics-informed computational framework that combines a Polarizable Unperturbed Electrostatic Potential (P-UEP) model with machine-learning-guided optimization and symmetry-driven relaxation to predict energetically favorable muon stopping sites in crystalline solids. MEOWN explicitly incorporates electrostatic interactions, electronic screening, polarization effects, and zero-point motion into the learning workflow, providing physically interpretable predictions with substantially reduced computational cost while maintaining physical consistency and scientific rigor. To validate the framework, we investigated several well-established benchmark materials, including MnSi, CoF2, CaF2, LiF, and NaF. The predicted muon stopping sites and dipolar fields are in close agreement with previously reported DFT+mu results, demonstrating the reliability and transferability of the approach across chemically diverse systems. Notably, MEOWN predicts the equilibrium muon stopping site within a few minutes, offering a significant speedup over conventional DFT-based calculations. This manuscript presents the theoretical foundations, computational methodology, and validation of MEOWN as a general-purpose software for rapid and reliable muon stopping-site prediction in crystalline materials.

发表机构

• Rajiv Gandhi Institute of Petroleum Technology（拉吉夫·甘地石油技术学院）

机构由 AI 辅助整理，请以论文原文为准。