利用伪拟势和几何递推构造线性偏微分方程的拟势
Construction of the quasi-potential for linear SDEs using false quasi-potentials and a geometric recursion
摘要:
拟势是随机微分方程大偏差理论中的一个关键概念。一旦找到关于相应确定性系统的吸引子的准势,就可以容易地获得最大似然退出路径,并从吸引吸引区域和吸引子附近的不变概率密度估计退出率。在原点具有渐近稳定平衡点的线性微分方程的拟势是一个二次型,其矩阵由Z. Chen和M. Freidlin(2005).他们的公式涉及某些矩阵指数的积分,并且不便于数值计算。在这项工作中,我提出了一种不同的方法来构建准势矩阵的线性偏微分方程的基础上,一定的易于获得的正交分解的启发所使用的M。Tao(2018).这些分解的一些性质,命名为假拟势分解,建立。给出了实现该方法的几何递归算法。它只涉及数值稳定的子算法,如计算舒尔分解和解决系统的线性方程或二次最小二乘问题。这项工作的动机是需要初始化的准潜在的求解器附近的渐近稳定的平衡。一种精确的方法是在平衡点附近线性化系统的势能,并将相邻网格点的准势能值设置为线性化系统的精确准势能。
英文摘要:
The quasi-potential is a key concept of the Large Deviation Theory for Stochastic Differential Equations (SDEs). Once the quasi-potential with respect to an attractor of the corresponding deterministic system is found, one can readily obtain maximum likelihood exit paths and estimate the exit rate from the basin of attraction and the invariant probability density near the attractor. The quasi-potential for a linear SDE with asymptotically stable equilibrium at the origin is a quadratic form whose matrix was found by Z. Chen and M. Freidlin (2005). Their formula involves an integral of certain matrix exponentials and is inconvenient for numerical evaluation. In this work, I propose a different approach for constructing the quasi-potential matrix for linear SDEs based on a certain easy-to-obtain hierarchy of orthogonal decompositions inspired by those used by M. Tao (2018). A number of properties of these decompositions, named false quasi-potential decompositions, is established. A geometric recursive algorithm implementing this approach is presented. It involves only numerically stable sub-algorithms such as the computation of Schur decompositions and solving systems of linear equations or quadratic least squares problems. This work is motivated by the need to initialize the quasi-potential solvers near asymptotically stable equilibria. An accurate way to do it is to linearize the SDE near the equilibrium and set the values of the quasi-potential at the neighboring mesh points to the exact quasi-potential of the linearized system.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。