NYU-ECNU , Center for Computational Chemistry at NYU, Shanghai , Shanghai 200062 , China.
J Chem Theory Comput. 2019 Apr 9;15(4):2195-2205. doi: 10.1021/acs.jctc.9b00072. Epub 2019 Mar 20.
In order to accelerate molecular dynamics simulations using polarizable force fields, we combine a new extended Lagrangian approach that eliminates the self-consistent field step (iEL/0-SCF) with a stochastic integration scheme that allows for a long time step using a multiple time scale algorithm (SIN(R)). We consider different algorithms for the combined scheme that places different components of the nonbonded forces into different time scales, as well as splitting individual nonbonded forces across time scales, to demonstrate that the combined method works well for bulk water as well as for a concentrated salt solution, aqueous peptide, and solvated protein. Depending on system and desired accuracy, the iEL/0-SCF and SIN(R) combination yields lower bound computational speed-ups of ∼6-8 relative to a molecular dynamics Verlet integration using a standard SCF solver implemented in the reference program TINKER 8.1. The combined approach embodies a significant advance for equilibrium simulations in the canonical ensemble of many-body potential energy surfaces for condensed phase systems with speed-ups that exceed what is possible by either method alone.
为了加速使用极化力场的分子动力学模拟,我们结合了一种新的扩展拉格朗日方法,该方法消除了自洽场步骤(iEL/0-SCF),并采用允许使用多时间尺度算法(SIN(R))的随机积分方案。我们考虑了组合方案的不同算法,这些算法将非键力的不同分量放入不同的时间尺度,以及将单个非键力跨时间尺度拆分,以证明该组合方法在水以及浓缩盐溶液、水性肽和溶剂化蛋白质中都能很好地工作。根据系统和所需的精度,iEL/0-SCF 和 SIN(R) 组合相对于使用标准 SCF 求解器在参考程序 TINKER 8.1 中实现的分子动力学 Verlet 积分,可实现约 6-8 的计算速度提高下限。该组合方法体现了在凝聚相系统的许多体势能表面的正则系综中进行平衡模拟的重大进展,其加速速度超过了任何单一方法都可能实现的速度。