A Positivity-Preserving Numerical Method for the Free Cox-Ingersoll-Ross Equation

Abstract

We introduce a positivity-preserving drift-implicit square-root Euler method for the free Cox--Ingersoll--Ross (CIR) equation driven by free Brownian motion. The method is obtained by discretizing the square-root equation implicitly in the drift and preserves strict positivity of the numerical solution without any step-size restriction. Under the Feller condition, we establish the uniqueness of the global strictly positive solution and derive an upper bound for its operator norm. A key ingredient for the error analysis is a free stochastic differential representation of the inverse square-root process. Another key ingredient is an a priori grid-point error estimate for sufficiently small step sizes that yields uniform upper and strictly positive lower spectral bounds for the numerical solution on any fixed finite time interval. These results support the strong convergence of order $1$ and improve the classical results. Moreover, both the exact and numerical solutions are exponentially stable with respect to perturbations of the initial value. The numerical stability holds for arbitrary step sizes, and the associated discrete decay rate converges to its continuous counterpart as the step size tends to zero. Numerical experiments confirm the theoretical results.