基于机器学习方法对非晶固体本征振动性质的研究

The understanding of the complex and anomalous vibrational properties of amorphous solids and glassy materials, especially for the nature of the boson peaks, has become one of the greatest challenges in the condensed matter physics. As the localized flow defects susceptible to rearrangement in disordered solids are particularly important in the scattering of sound waves, they are long considered to be closely related to the formation of the boson peak. Nevertheless, to identify the local ‘soft’ regime directly from the structure is extremely difficult, due to the microscopic particle-scale details largely missing and the disturbances of various complex factors as well, but the rapid development of the artificial intelligence and machine-learning methods in the recent years provides a new route to it. In our previous studies, we have showed a strong connection between the boson peak and the non-affine shear modulus for the first time in an experiment with an array of evidences. To further analyze the specific mechanisms of local structure and local mechanical characteristics in the formation of the boson peak, we will use machine-learning algorithms to classify the ‘soft’ and‘hard’ particles, accurately identifying both the structural and mechanical‘soft’ regime ranging from crystalline to the disordered solids. Moreover, the dependence of their statistical properties on the shapes of the boson peaks and the corresponding wavelengths will be analyzed in detail. Meanwhile, the commonality and differences of the mechanisms between the structural and mechanical ‘soft’ regimes on the anomalous vibrational properties will also be clarified.

非晶固体相比于晶体出现的反常振动性质，尤其对于玻色峰本质的理解，一直是当代凝聚态物理的一大难题。由于无序固体中易于发生塑性的局部缺陷结构在声波的散射过程中尤其重要，因而长期被认为与玻色峰的形成紧密相关。但由于在实际体系中缺乏对单粒子尺度微观细节的直接测量以及复杂因素的干扰，直接从结构上识别这种局部‘软’结构极其困难，而近年来人工智能及机器学习方法的快速发展为此提供了新途径。在前期的研究中，我们首次从实验上表明玻色峰与局域非仿射剪切模量有着很强的空间关联性。为进一步分析局部结构以及局部力学特征对玻色峰形成的具体作用机制，我们将运用机器学习算法对体系内部‘软’‘硬’颗粒进行自动分类，精确识别从二维光弹性颗粒晶体到无序体系的结构与力学‘软’区域，并详细分析其统计性质随玻色峰形状及相应波长的依赖关系，同时阐明二者对反常振动性质作用机制的共同点与差异。

玻璃及非晶固体相比于晶体存在着各种各样的反常振动性质，其中最典型的表现为其态密度相比于德拜模型有过剩态而形成了玻色峰，研究玻色峰形成的物理机制是理解非晶材料结构与热动力学行为的关键，但到目前为止并没有一个统一的定论。近年来，部分科学家通过各种理论研究表明玻色峰的形成与非晶体系中局部无序结构息息相关。. 本项目结合机器学习方法和光弹性实验技术，研究了不同无序度体系中结构与力学‘软’区域对二维非晶固体反常振动性质的影响。研究发现玻色峰形状与机器学习软度相关物理量之间满足单调依赖关系，并对局部‘软’区域如何影响玻色峰形状给出了物理解释；通过实验测量不同无序颗粒体系的剪切模量涨落，研究了玻色峰形成与负剪切模量所形成的局部力学软区域之间的关系，并证明基于剪切模量定义的局部力学软区域与局部无序结构共同决定了玻色峰的形状；通过对二维光弹性颗粒实验结果的研究与分析，证明了玻色峰与范霍夫奇点，以及玻色峰与安德森局域化之间的相互独立性；同时针对如何获取二维颗粒体系微观结构与力学信息等技术问题，课题组还对机器学习方法与光信息实验技术进行了深入探索。. 总之，课题组在已有的研究基础上，围绕不同无序度的非晶固体本征振动性质展开了多项模拟与实验研究，其研究成果对玻璃及非晶材料性质的进一步研究与应用具有一定的参考价值，对各地质灾害的理解与防治等都具有广泛的应用前景。