纳米材料的尺寸、结构和形貌等几何参数是决定其特殊物理和化学性能的关键因素，研究纳米结构的形成过程与机制对纳米科学的发展和纳米材料的广泛应用具有重要意义。传统结晶动力学理论认为颗粒的形成是通过单体逐渐累加完成的，其中包含成核和生长两个阶段，并相继形成了LaMer机理、Ostwald熟化、Lifshitz-Slyozov-Wagner机理和Watzky-Finke机理等著名的经典结晶理论。尽管经典成核-生长理论取得了巨大的成功，但是随着越来越多的纳米结构被制备出来，人们发现很多实验现象无法通过经典结晶机制加以解释或预测。因此，人们提到其中非传统结晶模型，它是一种以纳米颗粒或团簇为单元通过聚并生长的结晶模式。由于生长单元和演化尺度的改变，传统结晶理论已经不再适用，人们据此提出了一些新的生长机制，如取向连接，并发展出一系列的检测技术用于表征这种非传统结晶的特点。但是，受检测仪器和技术的时间、空间分辨能力的限制，无法获取颗粒聚并的动态过程以及颗粒间相互作用的变化等关键信息，使得我们无法从实验上深入分析非传统结晶的内在机制。随着计算机的运算能力突飞猛进，尤其是超级计算机的出现以及各种模拟算法和模型的发展，使得我们可以借助模拟的方法来弥补实验研究的不足。近年来人们逐渐意识到动力学过程对材料形貌演化的重要性，发现材料生长过程中所涉及的反应过程和扩散过程是材料生长中最具代表性的两个动力学过程。所以本文以银纳米颗粒的聚集生长为例，采用布朗动力学和分子动力学相结合的方法，对纳米及亚纳米尺度下银颗粒的聚并过程进行动力学模拟，分别构建了能够跟踪微观尺度上颗粒间相互作用的计算模型以及能够跟踪分子/原子尺度上颗粒表面原子间相互作用的计算模型，系统地研究了颗粒的聚并过程以及影响该过程的主要因素。主要有以下几点发现与成果：（1）颗粒聚集模型的构建。模拟纳米颗粒在不同聚集和扩散条件下的聚集行为，考虑到计算尺度以及计算时间，选取纳米颗粒和亚纳米颗粒两个颗粒尺度，引入布朗动力学（颗粒尺度模拟方法）以及分子动力学方法（原子/分子尺度模拟方法），不仅实现对纳米颗粒聚集过程的实时观测，还可以定量描述溶剂粘度、溶液离子强度等外界因素对颗粒聚并过程的影响。纳米级别的颗粒聚集采用DLVO作用势，此作用势包含颗粒间的范德华吸引势以及双电层排斥势，能够很好地反映纳米颗粒在溶液中的相互作用。同时，为提高计算效率，所用参数均进行归一化处理，以节省计算时间。因经典的DLVO作用势仅适用于球形颗粒，不能充分体现颗粒聚集过程的取向连接过程，所以对于亚纳米级别的颗粒选用全原子模型。选用全原子模型的颗粒为多面体结构，具有晶面。原子间相互作用选用EAM多体势，同时鉴于单体结构的稳定性，选取正二十面体和截角八面体作为聚集单元，正二十面体颗粒表面全部由 (111) 晶面构成，用于研究各向同性单元的聚并规律；而截角八面体由相对面积各异的 (111) 和 (100) 晶面构成，用于考察不同晶面间的各向异性聚集过程。（2）基于DLVO作用势的布朗动力学模拟纳米尺度下颗粒间的聚并过程。探索颗粒自身尺寸、聚集（溶液离子强度、颗粒表面电势）以及扩散（溶剂粘度）等因素对颗粒聚集速率（单体颗粒随时间的演化规律）、聚集体生长速率（聚集体平均尺寸及最大尺寸随时间的演化规律）以及聚集体数目等的影响，并对颗粒聚并生长过程进行定性及定量分析。在相同环境下，粒径较大的颗粒在溶液中的随机运动较弱，传质较慢且颗粒聚集所需跨越的能势垒较高，颗粒间聚集难度较大；而小颗粒布朗运动更剧烈，传质较快，且小颗粒间聚集所需跨越的能势垒较低，因此小颗粒聚并的概率要大于大颗粒；由于大小颗粒聚并难易程度的不同，颗粒聚并所得聚集体的生长速率均随着颗粒直径的增大而降低。溶液的离子强度与颗粒的表面电势决定了颗粒聚并所需跨越能势垒的高低，因此本文通过溶液离子强度与颗粒表面电势的调控聚集快慢。颗粒表面电势越小或溶液离子强度越大（双电层越薄），颗粒聚集所需跨越的能势垒越低，即聚集作用越强，颗粒间有效碰撞概率增加，聚集速率增大，聚集体生长速率也随之增大，颗粒间的聚并越容易。因溶剂作用于运动颗粒的曳力正比于颗粒的尺寸，所以在尺度较大的体系中，粘度对颗粒聚并过程的影响更加显著，粘度增加，颗粒扩散率大为降低，颗粒间碰撞概率和聚集速率明显减小，同样聚集体生长速率也随之减小。（3）分子动力学模拟亚纳米颗粒聚集生长过程。通过对不同聚集单元（正二十面体和截角八面体）在纯水中聚集过程的观察，得知所有聚集过程都是先由棱边原子接触开始，棱边原子形成的“接触点”将两颗粒连接，颗粒之间形成转轴，通过转轴转动完成颗粒面与面之间的融合。在完全相同的模拟条件下，截角八面体聚集体内部所含单体个数远大于正二十面体，说明含有高能面 (100) 的截角八面体更容易聚集；但由正二十面体形成的聚集体对称性更好。通过聚集路径分析发现，造成这种聚集差异的主要原因是聚集单元的表面能，单元特定晶面面积越大，该晶面的表面能就越大，聚集后体系能量下降越多，颗粒间优先在此晶面上完成聚集。同时，通过调节体系温度和溶剂粘度改变聚集单元的扩散能力，发现当扩散慢时，颗粒在溶液中运动缓慢，颗粒间有足够的弛豫时间，从而形成完美的晶向匹配，聚并成较为规整的结构；当扩散加快时则相反，聚集体内部结构的无序性增加。（4）本文最后在纳米和亚纳米两个尺度颗粒聚集生长的模拟工作基础上，设想了通过耦合两个尺度对应的模型、方法，模拟建立统一的颗粒聚并生长计算模型：颗粒间距离较近时选用全原子模型算法，距离较远时则自动切换至DLVO模型。并从作用势的角度证实了耦合模型的可行性。;Since the geometric parameters (sizes, structures, morphologies and so on) of nanomaterials are the key factors to determine their unique physical and chemical properties, investigations on the processes and mechanisms of nanostructure formation are significant for the development of nanoscience and the wide applications of nanomaterials. The traditional kinetic theory of crystallization proposes that the particles are formed through monomer-by-monomer addition, in which two steps are included, nucleation and growth. A series of classical crystallization mechanisms have been built up, like LaMer mechanism, Ostwald ripening, Lifshitz-Slyozov-Wagner mechanism and Watzky-Finke mechanism. In spite of the great success of the nucleation-growth theory, it is found that more and more crystallization phenomena cannot be interpreted and predicted by those classical mechanisms accompanying with numerous nanostructures being produced. In contrast to the conventional models, another category of crystallization by nanoparticle/cluster aggregation have been identified, which is called non-classical crystallization. Therefore, the classical theory is no more suitable due to the changes in precursors and precursor scales, and some novel mechanisms, such as oriented attachment, has been put forward and lots of techniques are being developed to characterize the features of non-classical crystallization. However, owing to the limitations on the temporal and spatial resolution of apparatuses, the dynamic process of agglomeration and interaction between aggregation building blocks cannot be accessed and thus it is difficult to achieve the principles of non-classical crystallization in depth from experimental aspects. Fortunately, the fast development of computers, especially the emergence of super calculation centers, combined with simulation algorithm and models, enables us to compensate the experimental shortages by numerical protocols.Over decades, material scientists were aware of the significance of kinetic processes for shape evolution. It was realized that the reaction process and diffusion process involved in the growth of materials are the two most representative kinetic processes. In this thesis, a combination of Brownian dynamics based DLVO methods and molecular dynamics was employed to simulate the aggregation behaviors of silver nanoparticles on the sub-nano and nano scales. Numerical models were built up to trace the crystalline alignment on microscale and the particle-to-particle interaction on mesoscale, and a systematic investigation was carried out on that how particles aggregated and what the decisive factors were. The main findings and achievements are shown as follows:(1) Construction of the basic frameworks for particle aggregation models. With the consideration for the computational scale and time scale, two simulation scales, sub-nano and nano scales were chosen to simulate the aggregation behavior of particles in different aggregation-diffusion systems. To observe the aggregation process of nanoparticles in real-time and quantify the influence of external factors like solvent viscosity and ionic strength by using the Brownian dynamic method and molecular dynamic method based on atom/molecule scale. As for the aggregation on the nanoscale, the DLVO potential was applied. DLVO potential contains Van der Waals attractive potential and double electic layer repulsive potential, which is suitable for describing the interaction of nanoparticles in solution. Meanwhile, every concerned parameter was normalized by the units of length, time and mass to accelerate the computional speed. However, the classical DLVO potential is employed to deal with building blocks as spheres and it has limitations in reflecting the oriented attachment of particles. Thus, the all atom model at sub-nano scale was proposed. The particle of all-atom model is polyhedron and it can present the interaction of different facets as facets appeared. The interaction between atoms was calculated by the embedded atom method and considering the stability of particles during simulations, icosahedrons and truncated octahedrons were selected as the building blocks. Icosahedrons were applied to explore the aggregation rules between isotropic units due icosahedral particles surrounded by pure (111) facets, while truncated octahedrons with different proportions of (111) and (100) facets were used to study the effects of anisotropic units on aggregate structures.(2) Aggregation processes simulated by Brownian dynamics based DLVO method on the nanoscale. The effects of building block sizes, aggregation (solution ion strength, surface potentials) and diffusion (solvent viscosity) were concluded by analyzing the aggregation rate of particles (temporal evolution of single nanoparticle), growth rate of aggregates (temporal evolution of average sizes, maximum sizes) and aggregates number. It was found that the larger was the building block, the slower was the random movement, so the more difficult was it to overcome the agglomeration barrier. For small particles, the Brownian motion was quite vigorous, so the aggregation probability was much higher than that of big particles. Due to the decline of agglomeration degrees, average and maximum sizes decreased with increasing build block sizes. Simultaneously, the energy barrier also depended on the surface potential and solution ion strength. Higher surface potentials and smaller ion strength (thicker double layers) resulted in higher energy barriers and gave rise to lower collision probability between particles. Therefore, both aggregation rates and aggregate sizes were reduced. Moreover, since the friction force employed on particles by solvents was proportional to both the particle size and solvent viscosity, the viscous effects were more significant for larger particles. Collision probability, aggregation rates and consequent aggregate sizes were obviously diminished.(3) Aggregation processes simulated by molecular dynamics on the sub-nanoscale. Via capturing the aggregate evolution from icosahedral and truncated octahedral building blocks, it is found that the particle aggregation started with the contact between edge atoms. Centering on the contact point, two particles completed crystalline alignment by free rotation and fusion of matched facets. With regard to agglomeration degrees, truncated octahedrons aggregated together more easily than icosahedrons, and truncated octahedral aggregates contained more building blocks and thus had bigger sizes, while icosahedral aggregates possessed more symmetrically geometric shapes. Through analysis of the aggregation pathways, the main reason accounting for above distinctions was the differences of surface energy, which depended on the surface areas of specific facets. Attachment preferred to occurred on those facets with large areas for that process was more thermodynamically favored because of more reduction of system energy after aggregation. In addition, the diffusive effects of building blocks on aggregation behaviors were investigated by adjusting the system temperature and solvent viscosity. The results indicated that ordered structures were formed under slow diffusion conditions because the random movement of building blocks was limited and they had enough relaxation time to perform lattice matching; otherwise, fast diffusion led to the increase of disorderness in aggregates.(4) Based on the simulations of particle aggregation on two scales, it is proposed that the future of the work in this thesis would be a combination of all-atom simulation and Brownian dynamics simulation into a consistently-formulated mesoscale model, in which the all-atom algorithm would be adopted when particles are close to each other, while the model would be switched into DLVO automatically when particles are mutually far away. And the feasibility of such a coupled mesoscale model would be demonstrated by calculating the interaction potential.