Stability of weak numerical schemes for stochastic differential equations (Q5906613)

A definition of numerical stability is proposed for stochastic differential equations with multiplicative noise. A class of Stratonovich type test equations is chosen that are of the form \[ dX_ t = (1 - \alpha) \lambda X_ t dt + \sqrt {\alpha}^ \gamma X_ t \circ dW_ t, \] where \(0 \leq \alpha \leq 2\), \(W\) is a real standard Wiener process and the parameters are as follows: \(\lambda = \lambda_ 1 + i \lambda_ 2\), \(\gamma = \gamma_ 1 + i \gamma_ 2\), \(\gamma_ 1 = - \sqrt {{1\over 2} (| \lambda | + \lambda_ 1)}\) and \(\gamma_ 2 = {-\lambda_ 2 \over 2 \gamma_ 1}\). Using these test equations, the regions of absolute stability are determined for different choices of the parameter \(\alpha\). These regions are found and depicted graphically for the explicit Euler scheme, the drift implicit Euler scheme, the fully implicit Euler scheme, and two symmetrical implicit Euler schemes. Although none are \(A\)-stable for all \(\alpha \in [0,2]\), the fully implicit Euler scheme is deemed preferable because of its large stability regions for \(\alpha\) near 0 and \(\alpha\) near 2.