Rounding Error Using Low Precision Approximate Random Variables

SIAM Journal on Scientific Computing, Volume 46, Issue 4, Page B502-B526, August 2024.
Abstract. For numerical approximations to stochastic differential equations using the Euler–Maruyama scheme, we propose incorporating approximate random variables computed using low precisions, such as single and half precision. We propose and justify a model for the rounding error incurred, and produce an average case error bound for two and four way differences, appropriate for regular and nested multilevel Monte Carlo estimations. Our rounding error model recovers and extends the statistical model by Arciniega and Allen [Stoch. Anal. Appl., 21 (2003), pp. 281–300], while bounding the size that systematic and biased rounding errors are permitted to be. By considering the variance structure of multilevel Monte Carlo correction terms in various precisions with and without a Kahan compensated summation, we compute the potential speed ups offered from the various precisions. We find single precision offers the potential for approximate speed improvements by a factor of 7 across a wide span of discretization levels. Half precision offers comparable improvements for several levels of coarse simulations, and even offers improvements by a factor of 10–12 for the very coarsest few levels, which are likely to dominate higher order methods such as the Milstein scheme.