Derandomization of the Euler scheme for scalar stochastic differential equations (Q413464): Difference between revisions

The authors consider a scalar stochastic differential equation with solution process \(X\), drift coefficient \(a\), diffusion coefficient \(b\), and initial value \(x\). They present an algorithm \(\hat{S}\) that computes a discrete distribution \(\hat{S}(x,a,b)= \sum_{y\in G\cup\{x\}} (Q^m)_{x,y}\,\delta_y\) as an approximation to the distribution \(S(x,a,b)\) of \(X(1)\). Here \(G\) is a set of equidistant nodes, \(Q\) is a transition matrix to at most six possible positions and \(m\) is the number of steps. A worst case analysis is provided for the computational cost and the error, assuming that the coefficients \(a\) and \(b\) have bounded derivatives up to order four. In terms of the computational cost the error is almost of the order 2/3 if the diffusion coefficient \(b\) is bounded away from zero, and of the order 1/2 in general.