液体在一定条件下可降温至凝固点以下而不发生结晶，此时的液体称为过冷液体。流动的过冷液体在自然界、日常生活以及许多工业领域中广泛存在。过冷液体往往具有低温（过冷）、高黏度、动态不均匀性等特征，呈现出复杂的流动性质。对过冷液体施加外剪切，随着剪切率升高，过冷液体会进入非线性区，其黏度（η）会随着剪切率（γ?）的升高而下降，并满足幂律关系η~γ?^(-λ)，称为剪切变稀。剪切变稀是过冷液体最重要、最常见的非线性流变学现象。研究过冷液体以剪切变稀为代表的非线性流变行为具有重要的理论与应用价值。论文提出了一种理解过冷液体非线性流变行为的微观物理图像。论文使用由实验确定的典型胶体过冷液体样品的相互作用势参数，开展了一系列布朗动力学稳态剪切模拟。通过分析模拟数据，论文指出剪切变稀源自某种局域弹性单元的演化过程，并将其命名为“局域弹性区”（localized elastic region, LER）。LER是过冷液体系统中暂时存在的区域，由中心粒子及其附近若干配位层的粒子构成。当存在外剪切时，LER内的粒子会作近似仿射运动，直至LER屈服。LER的特征应变γ和特征尺度ξ也满足与黏度类似的幂律关系γ~γ?^(ε)，ξ~γ?^(-ν)。这些幂指数满足一个简洁的标度率λ=1-ε=4ν，该式将系统非线性流变行为、弹性性质和微观构型关联起来。LER的松弛通过粒子沿剪切拉伸方向长度约为0.4粒子直径的非仿射跳跃位移实现。LER的弹性形变与屈服在流动中普遍存在，持续发生，并主导了系统的非线性黏弹性。论文使用流变小角中子散射，以胶体溶液为样品，对LER模型进行了实验研究。提出了一套从中子散射谱到LER特征量的数据处理方法，论证了流程中关键节点处理的数值有效性，定量验证了LER物理图像。为了在粒子层面进一步考察LER的微观结构特征和动态演化过程，论文使用机器学习方法研究了非线性剪切下过冷液体微观结构与动态之间的联系。通过对机器学习结果的分析，建立了LER与特定微观结构量的关联，进一步明确了LER的动态演化过程。
When lowering the temperature of a liquid, crystallization could be avoided under certain conditions even below the melting point. Such a liquid now is called a supercooled liquid. Flowing supercooled liquids are ubiquitous in nature, daily life, and a variety of industrial fields. Due to characteristics such as low temperature (being supercooled), high viscosity, and dynamic heterogeneity, supercooled liquids exhibit complex rheological behaviors. When subjected to large enough external shear, supercooled liquids would enter into a nonlinear regime termed shear thinning where the viscosity η decreases with increasing the shear rate γ?, expressed as η~γ?^(-λ). Shear thinning is the most important and common nonlinear rheological behavior of supercooled liquids. Studying the nonlinear rheological behaviors of supercooled liquids, especially shear thinning, is of fundamental and practical importance.This dissertation proposes a microscopic picture for understanding the nonlinear rheology of supercooled liquids. A series of steady shear Brownian dynamics simulations are conducted, using parameters determined from experiments of typical colloidal supercooled liquids. By analysing the simulation data, it is shown that the shear thinning of viscosity originates from the evolution of the localized elastic region (LER). An LER is a transient zone composed of the first several coordination shells of a reference particle. In response to the external shear, particles within LER undergo nearly affine displacement before the yielding of LER. The characteristic strain (γ) and size (ξ) of LER respectively depend on the shear rate by γ~γ?^(ε) and ξ~γ?^(-ν). Three exponents, λ, ε, and ν, are related by λ=1-ε=4ν. This simple relation connects the nonlinear rheology to the elastic properties and the microscopic configurational distortion of the system. The relaxation of the LER is promoted by the large-step nonaffine particle displacement along the extensional direction of the shear geometry with the step length of 0.4 particle diameter. The elastic deformation and relaxation of the LER are ubiquitous and successive in the flow, which compose the fundamental process governing the bulk nonlinear viscoelasticity.To experimentally study the LER model, this dissertation employs the rheological small-angle neutron scattering technique to study a set of supercooled colloidal suspensions. A comprehensive data reduction procedure is established, converting neutron scattering spectra into LER characteristic quantities. The numerical correctness of key steps in this procedure is validated. The experimental result well supports the LER picture.To further investigate the microscopic structural features and the dynamical evolution of LER at the particle level, this dissertation uses machine learning to examine the relationship between microscopic structural features and dynamics for supercooled liquids in the nonlinear regime. Through the analysis of machine learning results, correlations between LER and specific microscopic structural features are established. Moreover, the dynamical evolution process of LER is further elucidated.
