In this paper we study from a numerical analysis perspective the FractionalStep Kinetic Monte Carlo (FS-KMC) algorithms proposed in [1] for the parallelsimulation of spatially distributed particle systems on a lattice. FS-KMC arefractional step algorithms with a time-stepping window $\Delta t$, and as suchthey are inherently partially asynchronous since there is no processorcommunication during the period $\Delta t$. In this contribution we primarilyfocus on the error analysis of FS-KMC algorithms as approximations ofconventional, serial kinetic Monte Carlo (KMC). A key aspect of our analysisrelies on emphasising a goal-oriented approach for suitably defined macroscopicobservables (e.g., density, energy, correlations, surface roughness), ratherthan focusing on strong topology estimates for individual trajectories. One of the key implications of our error analysis is that it allows us toaddress systematically the processor communication of different parallelizationstrategies for KMC by comparing their (partial) asynchrony, which in turn ismeasured by their respective fractional time step $\Delta t$ for a prescribederror tolerance.
展开▼
机译:在本文中,我们从数值分析的角度研究了[1]中提出的分数步动力学蒙特卡罗(FS-KMC)算法,用于并行模拟晶格上空间分布的粒子系统。 FS-KMC是具有时间步窗$ \ Delta t $的分数步算法,因此,由于在$ \ Delta t $期间没有处理器通信,因此它们固有地是部分异步的。在此贡献中,我们主要关注于FS-KMC算法的误差分析,该算法是常规串行动力学蒙特卡洛(KMC)的近似值。我们分析的关键方面在于强调针对适当定义的宏观可观察物(例如密度,能量,相关性,表面粗糙度)的面向目标的方法,而不是专注于针对单个轨迹的强大拓扑估计。错误分析的关键含义之一是,它允许我们通过比较KMC的(部分)异步,系统地解决KMC的不同并行化策略的处理器通信,而异步则由它们各自的分数时间步长$ \ Delta t $来衡量。公差。
展开▼