Implementation of Milstein Schemes for Stochastic Delay-Differential Equations with Arbitrary Fixed Delays

TL;DR

This paper introduces numerical methods for stochastic delay-differential equations with arbitrary fixed delays, enabling accurate simulations without the need for delays to align with a uniform mesh.

Contribution

It develops simulation techniques for SDDEs with indivisible delays, achieving strong convergence orders 1/2 and 1 using fixed and variable step sizes.

Findings

Schemes achieve theoretical convergence orders in numerical experiments.

Linear interpolation and augmented meshes enable handling of indivisible delays.

Simulation of delayed stochastic integrals extends existing methods.

Abstract

This paper develops methods for numerically solving stochastic delay-differential equations (SDDEs) with multiple fixed delays that do not align with a uniform time mesh. We focus on numerical schemes of strong convergence orders $1/2$ and $1$, such as the Euler--Maruyama and Milstein schemes, respectively. Although numerical schemes for SDDEs with delays $\tau_1,\ldots,\tau_K$ are theoretically established, their implementations require evaluations at both present times such as $t_n$, and also at delayed times such as $t_n-\tau_k$ and $t_n-\tau_l-\tau_k$. As a result, previous simulations of these schemes have been largely restricted to the case of divisible delays. We develop simulation techniques for the general case of indivisible delays where delayed times such as $t_n-\tau_k$ are not restricted to a uniform time mesh. To achieve order of convergence (OoC) $1/2$, we implement the schemes with a fixed step size while using linear interpolation to approximate delayed scheme values. To achieve OoC $1$, we construct an augmented time mesh that includes all time points required to evaluate the schemes, which necessitates using a varying step size. We also introduce a technique to simulate delayed iterated stochastic integrals on the augmented time mesh, by extending an established method from the divisible-delays setting. We then confirm that the numerical schemes achieve their theoretical convergence orders with computational examples.