Computationally Efficient Multiconfigurational Reactive Molecular Dynamics.
Computationally Efficient Multiconfigurational Reactive Molecular Dynamics.
复制标题
计算高效的多构型反应分子动力学。
DOI：
10.1021/ct3006437
复制
发表时间：
2012
影响因子：
5.5
通讯作者：
Voth,GregoryA
中科院分区：
文献类型：
--
作者：
Yamashita,Takefumi;Peng,Yuxing;Knight,Chris;Voth,GregoryA
这是一个计算要求很高的任务，明确模拟系统中的电子自由度，以观察感兴趣的化学转化，同时采样收敛统计特性所需的时间和长度尺度，从而减少由于初始条件，有限尺寸效应和有限采样造成的伪影。一种显著降低计算费用的解决方案是分子模型，其中粒子之间的有效相互作用控制系统的动力学。如果这些模型中的相互作用势被开发来再现从电子结构计算和/或从头算分子动力学模拟计算出的性质，那么人们可以以计算成本的一小部分来计算准确的性质。多构型算法将系统建模为几种化学键合拓扑的线性组合以模拟化学反应，有时也称为“多态”。这些算法通常利用已经在流行的分子动力学软件包中发现的能量和力计算，从而在不显著改变代码结构的情况下促进它们的实现。然而，每个模拟步骤的几个键合拓扑结构的能量和力的评估可能会导致差的计算效率，如果冗余没有有效地去除，特别是相对于计算的长程库仑相互作用。本文提出了精确的近似（有效的长程相互作用和由此产生的混合方法）和多程序并行化策略，有效地计算反应分子模拟中的静电相互作用。
It is a computationally demanding task to explicitly simulate the electronic degrees of freedom in a system to observe the chemical transformations of interest, while at the same time sampling the time and length scales required to converge statistical properties and thus reduce artifacts due to initial conditions, finite-size effects, and limited sampling. One solution that significantly reduces the computational expense consists of molecular models in which effective interactions between particles govern the dynamics of the system. If the interaction potentials in these models are developed to reproduce calculated properties from electronic structure calculations and/or ab initio molecular dynamics simulations, then one can calculate accurate properties at a fraction of the computational cost. Multiconfigurational algorithms model the system as a linear combination of several chemical bonding topologies to simulate chemical reactions, also sometimes referred to as “multistate”. These algorithms typically utilize energy and force calculations already found in popular molecular dynamics software packages, thus facilitating their implementation without significant changes to the structure of the code. However, the evaluation of energies and forces for several bonding topologies per simulation step can lead to poor computational efficiency if redundancy is not efficiently removed, particularly with respect to the calculation of long-ranged Coulombic interactions. This paper presents accurate approximations (effective long-range interaction and resulting hybrid methods) and multiple-program parallelization strategies for the efficient calculation of electrostatic interactions in reactive molecular simulations.