Large-scale influence of numerical noises as artificial stochastic disturbances on a sustained turbulence

TL;DR

This study demonstrates that tiny artificial numerical noises can significantly alter large-scale flow structures in turbulence simulations, emphasizing the importance of noise control for accurate modeling.

Contribution

The paper introduces the use of clean numerical simulation (CNS) to quantify the impact of numerical noise on turbulence, revealing its potential to cause large-scale flow deviations.

Findings

Numerical noises can lead to different flow regimes in turbulence simulations.

CNS effectively reduces numerical noise, providing a reliable benchmark.

Artificial stochastic disturbances significantly influence large-scale turbulence behavior.

Abstract

We investigate the large-scale influence of numerical noises as tiny artificial stochastic disturbances on a sustained turbulence. Using the two-dimensional (2D) turbulent Rayleigh-B\'enard (RB) convection as an example, we numerically solve the NS equations, separately, by means of a traditional algorithm with double precision (marked by RKwD) and the so-called clean numerical simulation (CNS). The numerical simulation given by the RKwD is a mixture of the "true" physical solution and the "false" numerical noises that is random and can be regarded as a kind of artificial stochastic disturbances: unfortunately, the "true" physical solution is mostly at the same level as the "false" numerical noises. By contrast, the CNS can greatly reduce the background numerical noise to any a required level so that the "false" numerical noises are negligible compared with the "true" physical solution and thus the CNS solution can be used as a "clean" benchmark solution for comparison. It is found that the numerical noises as tiny artificial stochastic disturbances could indeed lead to large-scale deviations of simulations not only in spatio-temporal trajectories but also even in statistics. Especially, these numerical noises (as artificial stochastic disturbances) even lead to different types of flows: the shearing convection occurs for the RKwD simulations, and its corresponding flow field turns to a kind of zonal flow thereafter, however the CNS benchmark solution always sustains the non-shearing vortical/roll-like convection during the whole process of simulation. Thus, we provide a rigorous evidence that numerical noises as a kind of small-scale artificial stochastic disturbances have quantitatively and qualitatively large-scale influences on a sustained turbulence, i.e. the 2D turbulent RB convection considered in this paper.