Algorithms for integration of stochastic differential equations using parallel optimized sampling in the Stratonovich calculus

详细信息    查看全文

• 作者：Simon Kiesewetter ; ^{skiesewetter@swin.edu.au} ; Peter D. Drummond
• 关键词：Stochastic ; Differential ; Equation ; Stratonovich ; Variance ; Reduction
• 刊名：Computer Physics Communications
• 出版年：2017
• 出版时间：March 2017
• 年：2017
• 卷：212
• 期：Complete
• 页码：25-38
• 全文大小：1843 K
• 卷排序：212

文摘

A variance reduction method for stochastic integration of Fokker–Planck equations is derived. This unifies the cumulant hierarchy and stochastic equation approaches to obtaining moments, giving a performance superior to either. We show that the brute force method of reducing sampling error by just using more trajectories in a sampled stochastic equation is not the best approach. The alternative of using a hierarchy of moment equations is also not optimal, as it may converge to erroneous answers. Instead, through Bayesian conditioning of the stochastic noise on the requirement that moment equations are satisfied, we obtain improved results with reduced sampling errors for a given number of stochastic trajectories. The method used here converges faster in time-step than Ito–Euler algorithms. This parallel optimized sampling (POS) algorithm is illustrated by several examples, including a bistable nonlinear oscillator case where moment hierarchies fail to converge.