Rudzinski Joseph F, Noid William G
Department of Chemistry, The Pennsylvania State University , University Park, Pennsylvania 16802, United States.
J Phys Chem B. 2014 Jul 17;118(28):8295-312. doi: 10.1021/jp501694z. Epub 2014 Apr 16.
Low resolution coarse-grained (CG) models enable highly efficient simulations of complex systems. The interactions in CG models are often iteratively refined over multiple simulations until they reproduce the one-dimensional (1-D) distribution functions, e.g., radial distribution functions (rdfs), of an all-atom (AA) model. In contrast, the multiscale coarse-graining (MS-CG) method employs a generalized Yvon-Born-Green (g-YBG) relation to determine CG potentials directly (i.e., without iteration) from the correlations observed for the AA model. However, MS-CG models do not necessarily reproduce the 1-D distribution functions of the AA model. Consequently, recent studies have incorporated the g-YBG equation into iterative methods for more accurately reproducing AA rdfs. In this work, we consider a theoretical framework for an iterative g-YBG method. We numerically demonstrate that the method robustly determines accurate models for both hexane and also a more complex molecule, 3-hexylthiophene. By examining the MS-CG and iterative g-YBG models for several distinct CG representations of both molecules, we investigate the approximations of the MS-CG method and their sensitivity to the CG mapping. More generally, we explicitly demonstrate that CG models often reproduce 1-D distribution functions of AA models at the expense of distorting the cross-correlations between the corresponding degrees of freedom. In particular, CG models that accurately reproduce intramolecular 1-D distribution functions may still provide a poor description of the molecular conformations sampled by the AA model. We demonstrate a simple and predictive analysis for determining CG mappings that promote an accurate description of these molecular conformations.
低分辨率粗粒化（CG）模型能够对复杂系统进行高效模拟。CG模型中的相互作用通常会在多次模拟中反复优化，直到它们能够重现全原子（AA）模型的一维（1-D）分布函数，例如径向分布函数（rdfs）。相比之下，多尺度粗粒化（MS-CG）方法采用广义的伊冯-博恩-格林（g-YBG）关系，直接（即无需迭代）根据AA模型中观察到的相关性来确定CG势。然而，MS-CG模型不一定能重现AA模型的1-D分布函数。因此，最近的研究将g-YBG方程纳入迭代方法，以更准确地重现AA的rdfs。在这项工作中，我们考虑了一种迭代g-YBG方法的理论框架。我们通过数值证明，该方法能够稳健地为己烷以及更复杂的分子3-己基噻吩确定准确的模型。通过研究这两种分子几种不同CG表示的MS-CG和迭代g-YBG模型，我们研究了MS-CG方法的近似性及其对CG映射的敏感性。更一般地说，我们明确证明，CG模型通常以扭曲相应自由度之间的交叉相关性为代价来重现AA模型的1-D分布函数。特别是，能够准确重现分子内1-D分布函数的CG模型，对于AA模型所采样的分子构象可能仍然提供较差的描述。我们展示了一种简单且具有预测性的分析方法，用于确定能够促进对这些分子构象进行准确描述的CG映射。
