Large scale simulation of macromolecules in solution: Combining the periodic fast multipole method with multiple time step integrators
Large scale simulation of macromolecules in solution: Combining the periodic fast multipole method with multiple time step integrators
复制标题
DOI：
10.1063/1.474115
复制
发表时间：
1997-06-15
影响因子：
4.4
通讯作者：
Berne, BJ
中科院分区：
文献类型：
--
作者：
Figueirido, F;Levy, RM;Berne, BJ
对溶液中的大分子进行大规模模拟变得可行，而不会受到力截断产生的伪影的影响。快速多极子法（FMM）等新型力评估算法与可逆参考系统传播器算法（r-RESPA）等多时间尺度积分方法相结合，用于对大分子系统进行快速稳定的模拟。在具有周期性边界条件的模拟中对远程力的一致处理需要使用周期性形式的库仑势。本文将FMM扩展到周期系统，并与RESPA相结合，产生了一种新算法，并成功应用于溶液中大生物分子的模拟。如果不同阶段的相互作用能够顺利分离，那么在超过 40 000 个原子的系统上，即使时间步长达到 12 fs，也能获得良好的能量守恒，并且与库仑相互作用的 Ewald 和的标准 Verlet 积分器相比，CPU 加速可达到 20 倍以上。与最近开发的粒子网格Ewald（PME）方法相比，周期性r-RESPA/FMM在大约20 000个原子处具有盈亏平衡点；对于较大的系统，r-RESPA/FMM 预计效率更高。 (C) 1997 年美国物理研究所。
Large scale simulations of macromolecules in solution that do not suffer from artifacts arising from force truncations are becoming feasible. New force evaluation algorithms such as the fast multipole method (FMM) and multiple time scale integration methods such as the reversible reference system propagator algorithm (r-RESPA) have been combined and used to perform fast and stable simulations of large macromolecular systems. A consistent treatment of the long-range forces in simulations with periodic boundary conditions requires the use of a periodic form of the Coulomb potential. In this article, the FMM is extended to periodic systems, and combined with RESPA, yielding a new algorithm that is successfully applied to the simulation of large biomolecules in solution. If the interactions at different stages are separated smoothly, good energy conservation is obtained even for time steps as large as 12 fs on a system of over 40 000 atoms, and a CPU speedup of more than a factor of 20 is achieved compared to the standard Verlet integrator with Ewald sum for the Coulombic interaction. As compared with the recently developed particle-mesh Ewald (PME) method, the periodic r-RESPA/FMM has a break-even point at about 20 000 atoms; for larger systems, r-RESPA/FMM is expected to be more efficient. (C) 1997 American Institute of Physics.